### Spoken transcript Skip to main contentSkip to footer Try Claude Engineering at Anthropic Effective context engineering for AI agents Published Sep 29, 2025 Context is a critical but finite resource for AI agents. In this post, we explore strategies for effectively curating and managing the context that powers them. After a few years of prompt engineering being the focus of attention in applied AI, a new term has come to prominence: context engineering. Building with language models is becoming less about finding the right words and phrases for your prompts, and more about answering the broader question of “what configuration of context is most likely to generate our model’s desired behavior?" Context refers to the set of tokens included when sampling from a large-language model (LLM). The engineering problem at hand is optimizing the utility of those tokens against the inherent constraints of LLMs in order to consistently achieve a desired outcome. Effectively wrangling LLMs often requires thinking in context — in other words: considering the holistic state available to the LLM at any given time and what potential behaviors that state might yield. In this post, we’ll explore the emerging art of context engineering and offer a refined mental model for building steerable, effective agents. Context engineering vs. prompt engineering At Anthropic, we view context engineering as the natural progression of prompt engineering. Prompt engineering refers to methods for writing and organizing LLM instructions for optimal outcomes (see our docs for an overview and useful prompt engineering strategies). Context engineering refers to the set of strategies for curating and maintaining the optimal set of tokens (information) during LLM inference, including all the other information that may land there outside of the prompts. In the early days of engineering with LLMs, prompting was the biggest component of AI engineering work, as the majority of use cases outside of everyday chat interactions required prompts optimized for one-shot classification or text generation tasks. As the term implies, the primary focus of prompt engineering is how to write effective prompts, particularly system prompts. However, as we move towards engineering more capable agents that operate over multiple turns of inference and longer time horizons, we need strategies for managing the entire context state (system instructions, tools, Model Context Protocol (MCP), external data, message history, etc). An agent running in a loop generates more and more data that could be relevant for the next turn of inference, and this information must be cyclically refined. Context engineering is the art and science of curating what will go into the limited context window from that constantly evolving universe of possible information. In contrast to the discrete task of writing a prompt, context engineering is iterative and the curation phase happens each time we decide what to pass to the model. Why context engineering is important to building capable agents Despite their speed and ability to manage larger and larger volumes of data, we’ve observed that LLMs, like humans, lose focus or experience confusion at a certain point. Studies on needle-in-a-haystackstyle benchmarking have uncovered the concept of context rot: as the number of tokens in the context window increases, the model’s ability to accurately recall information from that context decreases. While some models exhibit more gentle degradation than others, this characteristic emerges across all models. Context, therefore, must be treated as a finite resource with diminishing marginal returns. Like humans, who have limited working memory capacity, LLMs have an “attention budget” that they draw on when parsing large volumes of context. Every new token introduced depletes this budget by some amount, increasing the need to carefully curate the tokens available to the LLM. This attention scarcity stems from architectural constraints of LLMs. LLMs are based on the transformer architecture, which enables every token to attend to every other token across the entire context. This results in n² pairwise relationships for n tokens. As its context length increases, a model's ability to capture these pairwise relationships gets stretched thin, creating a natural tension between context size and attention focus. Additionally, models develop their attention patterns from training data distributions where shorter sequences are typically more common than longer ones. This means models have less experience with, and fewer specialized parameters for, context-wide dependencies. Techniques like position encoding interpolation allow models to handle longer sequences by adapting them to the originally trained smaller context, though with some degradation in token position understanding. These factors create a performance gradient rather than a hard cliff: models remain highly capable at longer contexts but may show reduced precision for information retrieval and long-range reasoning compared to their performance on shorter contexts. These realities mean that thoughtful context engineering is essential for building capable agents. The anatomy of effective context Given that LLMs are constrained by a finite attention budget, good context engineering means finding the smallestpossible set of high-signal tokens that maximize the likelihood of some desired outcome. Implementing this practice is much easier said than done, but in the following section, we outline what this guiding principle means in practice across the different components of context. System prompts should be extremely clear and use simple, direct language that presents ideas at the right altitude for the agent. The right altitude is the Goldilocks zone between two common failure modes. At one extreme, we see engineers hardcoding complex, brittle logic in their prompts to elicit exact agentic behavior. This approach creates fragility and increases maintenance complexity over time. At the other extreme, engineers sometimes provide vague, high-level guidance that fails to give the LLM concrete signals for desired outputs or falsely assumes shared context. The optimal altitude strikes a balance: specific enough to guide behavior effectively, yet flexible enough to provide the model with strong heuristics to guide behavior. At one end of the spectrum, we see brittle if-else hardcoded prompts, and at the other end we see prompts that are overly general or falsely assume shared context. We recommend organizing prompts into distinct sections (like <background_information>, <instructions>, ## Tool guidance, ## Output description, etc) and using techniques like XML tagging or Markdown headers to delineate these sections, although the exact formatting of prompts is likely becoming less important as models become more capable. Regardless of how you decide to structure your system prompt, you should be striving for the minimal set of information that fully outlines your expected behavior. (Note that minimal does not necessarily mean short; you still need to give the agent sufficient information up front to ensure it adheres to the desired behavior.) It’s best to start by testing a minimal prompt with the best model available to see how it performs on your task, and then add clear instructions and examples to improve performance based on failure modes found during initial testing. Tools allow agents to operate with their environment and pull in new, additional context as they work. Because tools define the contract between agents and their information/action space, it’s extremely important that tools promote efficiency, both by returning information that is token efficient and by encouraging efficient agent behaviors. In Writing tools for AI agents – with AI agents, we discussed building tools that are well understood by LLMs and have minimal overlap in functionality. Similar to the functions of a well-designed codebase, tools should be self-contained, robust to error, and extremely clear with respect to their intended use. Input parameters should similarly be descriptive, unambiguous, and play to the inherent strengths of the model. One of the most common failure modes we see is bloated tool sets that cover too much functionality or lead to ambiguous decision points about which tool to use. If a human engineer can’t definitively say which tool should be used in a given situation, an AI agent can’t be expected to do better. As we’ll discuss later, curating a minimal viable set of tools for the agent can also lead to more reliable maintenance and pruning of context over long interactions. Providing examples, otherwise known as few-shot prompting, is a well known best practice that we continue to strongly advise. However, teams will often stuff a laundry list of edge cases into a prompt in an attempt to articulate every possible rule the LLM should follow for a particular task. We do not recommend this. Instead, we recommend working to curate a set of diverse, canonical examples that effectively portray the expected behavior of the agent. For an LLM, examples are the “pictures” worth a thousand words. Our overall guidance across the different components of context (system prompts, tools, examples, message history, etc) is to be thoughtful and keep your context informative, yet tight. Now let's dive into dynamically retrieving context at runtime. Context retrieval and agentic search In Building effective AI agents, we highlighted the differences between LLM-based workflows and agents. Since we wrote that post, we’ve gravitated towards a simple definition for agents: LLMs autonomously using tools in a loop. Working alongside our customers, we’ve seen the field converging on this simple paradigm. As the underlying models become more capable, the level of autonomy of agents can scale: smarter models allow agents to independently navigate nuanced problem spaces and recover from errors. We’re now seeing a shift in how engineers think about designing context for agents. Today, many AI-native applications employ some form of embedding-based pre-inference time retrieval to surface important context for the agent to reason over. As the field transitions to more agentic approaches, we increasingly see teams augmenting these retrieval systems with “just in time” context strategies. Rather than pre-processing all relevant data up front, agents built with the “just in time” approach maintain lightweight identifiers (file paths, stored queries, web links, etc.) and use these references to dynamically load data into context at runtime using tools. Anthropic’s agentic coding solution Claude Code uses this approach to perform complex data analysis over large databases. The model can write targeted queries, store results, and leverage Bash commands like head and tail to analyze large volumes of data without ever loading the full data objects into context. This approach mirrors human cognition: we generally don’t memorize entire corpuses of information, but rather introduce external organization and indexing systems like file systems, inboxes, and bookmarks to retrieve relevant information on demand. Beyond storage efficiency, the metadata of these references provides a mechanism to efficiently refine behavior, whether explicitly provided or intuitive. To an agent operating in a file system, the presence of a file named test_utils.py in a tests folder implies a different purpose than a file with the same name located in src/core_logic/ Folder hierarchies, naming conventions, and timestamps all provide important signals that help both humans and agents understand how and when to utilize information. Letting agents navigate and retrieve data autonomously also enables progressive disclosure—in other words, allows agents to incrementally discover relevant context through exploration. Each interaction yields context that informs the next decision: file sizes suggest complexity; naming conventions hint at purpose; timestamps can be a proxy for relevance. Agents can assemble understanding layer by layer, maintaining only what's necessary in working memory and leveraging note-taking strategies for additional persistence. This self-managed context window keeps the agent focused on relevant subsets rather than drowning in exhaustive but potentially irrelevant information. Of course, there's a trade-off: runtime exploration is slower than retrieving pre-computed data. Not only that, but opinionated and thoughtful engineering is required to ensure that an LLM has the right tools and heuristics for effectively navigating its information landscape. Without proper guidance, an agent can waste context by misusing tools, chasing dead-ends, or failing to identify key information. In certain settings, the most effective agents might employ a hybrid strategy, retrieving some data up front for speed, and pursuing further autonomous exploration at its discretion. The decision boundary for the ‘right’ level of autonomy depends on the task. Claude Code is an agent that employs this hybrid model: CLAUDE.md files are naively dropped into context up front, while primitives like glob and grep allow it to navigate its environment and retrieve files just-in-time, effectively bypassing the issues of stale indexing and complex syntax trees. The hybrid strategy might be better suited for contexts with less dynamic content, such as legal or finance work. As model capabilities improve, agentic design will trend towards letting intelligent models act intelligently, with progressively less human curation. Given the rapid pace of progress in the field, "do the simplest thing that works" will likely remain our best advice for teams building agents on top of Claude. Context engineering for long-horizon tasks Long-horizon tasks require agents to maintain coherence, context, and goal-directed behavior over sequences of actions where the token count exceeds the LLM’s context window. For tasks that span tens of minutes to multiple hours of continuous work, like large codebase migrations or comprehensive research projects, agents require specialized techniques to work around the context window size limitation. Waiting for larger context windows might seem like an obvious tactic. But it's likely that for the foreseeable future, context windows of all sizes will be subject to context pollution and information relevance concerns—at least for situations where the strongest agent performance is desired. To enable agents to work effectively across extended time horizons, we've developed a few techniques that address these context pollution constraints directly: compaction, structured note-taking, and multi-agent architectures. Compaction Compaction is the practice of taking a conversation nearing the context window limit, summarizing its contents, and reinitiating a new context window with the summary. Compaction typically serves as the first lever in context engineering to drive better long-term coherence. At its core, compaction distills the contents of a context window in a high-fidelity manner, enabling the agent to continue with minimal performance degradation. In Claude Code, for example, we implement this by passing the message history to the model to summarize and compress the most critical details. The model preserves architectural decisions, unresolved bugs, and implementation details while discarding redundant tool outputs or messages. The agent can then continue with this compressed context plus the five most recently accessed files. Users get continuity without worrying about context window limitations. The art of compaction lies in the selection of what to keep versus what to discard, as overly aggressive compaction can result in the loss of subtle but critical context whose importance only becomes apparent later. For engineers implementing compaction systems, we recommend carefully tuning your prompt on complex agent traces. Start by maximizing recall to ensure your compaction prompt captures every relevant piece of information from the trace, then iterate to improve precision by eliminating superfluous content. An example of low-hanging superfluous content is clearing tool calls and results – once a tool has been called deep in the message history, why would the agent need to see the raw result again? One of the safest lightest touch forms of compaction is tool result clearing, most recently launched as a feature on the Claude Developer Platform. Structured note-taking Structured note-taking, or agentic memory, is a technique where the agent regularly writes notes persisted to memory outside of the context window. These notes get pulled back into the context window at later times. This strategy provides persistent memory with minimal overhead. Like Claude Code creating a to-do list, or your custom agent maintaining a NOTES.md file, this simple pattern allows the agent to track progress across complex tasks, maintaining critical context and dependencies that would otherwise be lost across dozens of tool calls. Claude playing Pokémon demonstrates how memory transforms agent capabilities in non-coding domains. The agent maintains precise tallies across thousands of game steps—tracking objectives like "for the last 1,234 steps I've been training my Pokémon in Route 1, Pikachu has gained 8 levels toward the target of 10." Without any prompting about memory structure, it develops maps of explored regions, remembers which key achievements it has unlocked, and maintains strategic notes of combat strategies that help it learn which attacks work best against different opponents. After context resets, the agent reads its own notes and continues multi-hour training sequences or dungeon explorations. This coherence across summarization steps enables long-horizon strategies that would be impossible when keeping all the information in the LLM’s context window alone. As part of our Sonnet 4.5 launch, we released a memory tool in public beta on the Claude Developer Platform that makes it easier to store and consult information outside the context window through a file-based system. This allows agents to build up knowledge bases over time, maintain project state across sessions, and reference previous work without keeping everything in context. Sub-agent architectures Sub-agent architectures provide another way around context limitations. Rather than one agent attempting to maintain state across an entire project, specialized sub-agents can handle focused tasks with clean context windows. The main agent coordinates with a high-level plan while subagents perform deep technical work or use tools to find relevant information. Each subagent might explore extensively, using tens of thousands of tokens or more, but returns only a condensed, distilled summary of its work (often 1,000-2,000 tokens). This approach achieves a clear separation of concerns—the detailed search context remains isolated within sub-agents, while the lead agent focuses on synthesizing and analyzing the results. This pattern, discussed in How we built our multi-agent research system, showed a substantial improvement over single-agent systems on complex research tasks. The choice between these approaches depends on task characteristics. For example: Compaction maintains conversational flow for tasks requiring extensive back-and-forth; Note-taking excels for iterative development with clear milestones; Multi-agent architectures handle complex research and analysis where parallel exploration pays dividends. Even as models continue to improve, the challenge of maintaining coherence across extended interactions will remain central to building more effective agents. Conclusion Context engineering represents a fundamental shift in how we build with LLMs. As models become more capable, the challenge isn't just crafting the perfect prompt—it's thoughtfully curating what information enters the model's limited attention budget at each step. Whether you're implementing compaction for long-horizon tasks, designing token-efficient tools, or enabling agents to explore their environment just-in-time, the guiding principle remains the same: find the smallest set of high-signal tokens that maximize the likelihood of your desired outcome. The techniques we've outlined will continue evolving as models improve. We're already seeing that smarter models require less prescriptive engineering, allowing agents to operate with more autonomy. But even as capabilities scale, treating context as a precious, finite resource will remain central to building reliable, effective agents. Get started with context engineering in the Claude Developer Platform today, and access helpful tips and best practices via our memory and context management cookbook. Acknowledgements Written by Anthropic's Applied AI team: Prithvi Rajasekaran, Ethan Dixon, Carly Ryan, and Jeremy Hadfield, with contributions from team members Rafi Ayub, Hannah Moran, Cal Rueb, and Connor Jennings. Special thanks to Molly Vorwerck, Stuart Ritchie, and Maggie Vo for their support. Get the developer newsletter Product updates, how-tos, community spotlights, and more. Delivered monthly to your inbox. ### On-screen text ### Every link captured - Try Claude -> https://claude.ai/ - our docs -> https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/overview - Model Context Protocol -> https://modelcontextprotocol.io/docs/getting-started/intro - art and science -> https://x.com/karpathy/status/1937902205765607626?lang=en - context rot -> https://research.trychroma.com/context-rot - limited working memory capacity -> https://journals.sagepub.com/doi/abs/10.1177/0963721409359277 - transformer architecture -> https://arxiv.org/abs/1706.03762 - attend to every other token -> https://huggingface.co/blog/Esmail-AGumaan/attention-is-all-you-need - position encoding interpolation -> https://arxiv.org/pdf/2306.15595 - Writing tools for AI agents – with AI agents -> https://www.anthropic.com/engineering/writing-tools-for-agents - Building effective AI agents -> https://www.anthropic.com/research/building-effective-agents - simple definition -> https://simonwillison.net/2025/Sep/18/agents/ - Claude Code -> https://www.anthropic.com/claude-code - CLAUDE.md -> http://claude.md - feature on the Claude Developer Platform -> https://www.anthropic.com/news/context-management - Claude playing Pokémon -> https://www.twitch.tv/claudeplayspokemon - Sonnet 4.5 launch -> https://www.anthropic.com/effective-context-engineering-for-ai-agents - a memory tool -> http://anthropic.com/news/context-management - How we built our multi-agent research system -> https://www.anthropic.com/engineering/multi-agent-research-system - memory and context management -> https://platform.claude.com/cookbook/tool-use-memory-cookbook - Claude -> https://claude.com/product/overview - Claude Code -> https://claude.com/product/claude-code - Claude Code Enterprise -> https://claude.com/product/claude-code/enterprise - Claude Cowork -> https://claude.com/product/cowork - @Claude -> https://claude.com/product/tag - Claude Design -> https://claude.com/product/design - Claude Security -> https://claude.com/product/claude-security - Claude for Chrome -> https://claude.com/chrome - Claude for Microsoft 365 -> https://claude.com/claude-for-microsoft-365 - Skills -> https://www.claude.com/skills - Download app -> https://claude.ai/download - Pricing -> https://claude.com/pricing - Mythos -> https://www.anthropic.com/claude/mythos - Fable -> https://www.anthropic.com/claude/fable - Opus -> https://www.anthropic.com/claude/opus - Sonnet -> https://www.anthropic.com/claude/sonnet - Haiku -> https://www.anthropic.com/claude/haiku - AI agents -> https://claude.com/solutions/agents - Code modernization -> https://claude.com/solutions/code-modernization - Coding -> https://claude.com/solutions/coding - Customer support -> https://claude.com/solutions/customer-support - Education -> https://claude.com/solutions/education - Enterprise -> https://claude.com/solutions/enterprise - Financial services -> https://claude.com/solutions/financial-services - Government -> https://claude.com/solutions/government - Healthcare -> https://claude.com/solutions/healthcare - Legal -> https://claude.com/solutions/legal - Life sciences -> https://claude.com/solutions/life-sciences - Nonprofits -> https://claude.com/solutions/nonprofits - Security -> https://claude.com/solutions/security - Small business -> https://claude.com/solutions/small-business - Startups -> https://claude.com/programs/startups - Overview -> https://claude.com/platform/api - Developer docs -> https://platform.claude.com/docs - Pricing -> https://claude.com/pricing#api - Ecosystem -> https://claude.com/ecosystem - Marketplace -> https://claude.com/platform/marketplace - Regional compliance -> https://claude.com/regional-compliance - Claude on AWS -> https://claude.com/partners/claude-on-aws - Google Cloud -> https://claude.com/partners/google-cloud-vertex-ai - Microsoft Foundry -> https://claude.com/partners/microsoft-foundry - Console login -> https://platform.claude.com/ - Blog -> https://claude.com/blog - Claude partner network -> https://claude.com/partners - Community -> https://claude.com/community - Connectors -> https://claude.com/connectors - Customer stories -> https://claude.com/customers - Plugins -> https://claude.com/plugins - Powered by Claude -> https://claude.com/partners/powered-by-claude - Service partners -> https://claude.com/partners/services - Tutorials -> https://claude.com/resources/tutorials - Use cases -> https://claude.com/resources/use-cases - Availability -> https://www.anthropic.com/supported-countries - Status -> https://status.anthropic.com/ - Support center -> https://support.claude.com/en/ - Responsible Scaling Policy -> https://www.anthropic.com/news/announcing-our-updated-responsible-scaling-policy - Security and compliance -> https://trust.anthropic.com/ - Privacy policy -> https://www.anthropic.com/legal/privacy - Consumer health data privacy policy -> https://www.anthropic.com/legal/consumer-health-data-privacy-policy - Responsible disclosure policy -> https://www.anthropic.com/responsible-disclosure-policy - Terms of service: Commercial -> https://www.anthropic.com/legal/commercial-terms - Terms of service: Consumer -> https://www.anthropic.com/legal/consumer-terms - Usage policy -> https://www.anthropic.com/legal/aup - (link) -> https://www.linkedin.com/company/anthropicresearch - (link) -> https://x.com/AnthropicAI - (link) -> https://www.youtube.com/@anthropic-ai

### Spoken transcript I grew my newsletter from 500 people to over 10,000 in about four weeks. And the way I was able to do this is by putting out really high quality lead magnets, but not in the way you think. I build the guide in Notion, set up a beautiful landing page that requires people to put their email in to download it, and then an NNN workflow to deliver it. But the crazy part is that takes me 10 minutes to build. And I'm going to let you in on the secret. It's all thanks to Cloud Code. All right, start the timer. Okay, so we've got the digital products built out in Notion, and we're just going to go up here and grab the published share link. Then we're going to run the skill I created called the lead magnet launcher. Paste in our Notion URL and send it off. like what we're even promoting. It's created the landing page, building the NNN workflow, and now it's done. So it's gone and created this really, really clean looking landing page. And it's even set up the domain correctly. And it's live on the web. When you enter your email, it takes you to this page to download the playbook. And if we go into NNN, you can see it's created this workflow that delivers the lead magnet to the user via email. And that's a clawed skill, by the way. And I've got these skills built out for all sorts of repeatable processes like this. And they literally buy me back 40, 50 plus hours a week. Go make some skills right now. ### On-screen text grew my newsletter from 500 people Subscriber growth Last 4 weeks Unsubscribed • Posts sent in about and the way was able to do this putting out really high quality lead magnets you think toggie tri webhook list Once you ve do to focus or unfocu The OpenClaw Play.... Guide 1: How OpenClaw Ac Edited 2d ago Digital Products The OpenClaw Playbook The OpenClaw Playbook Everything from your first install to a fully autonomous Al team. Who This Is For You've heard about OpenClaw. Maybe you've already tried to set it up and got stuck. Maybe your is running but feels like a worse ChatGPT. Maybe you saw someone's multi-agent setup on Twitterand thoucht want that This guide series takes you from "what even is this?" to a fully configured Al assistant that remembers, learns, works on its own, and follows your specific business processes. build the guide shift option control command option command to youa) toggis the a Once you've de to focus or unfoc Build Your Clawdbot Org Chart: From One Bot to a Full Al Team erpent B- hours in the official OpenClaw aocs and turned it into a step-hy-sten guide. Foun multi-agent structures - trom a single bot to e tull nompany org chart with boss agents, specilist workers, and cross-team communication. set up a delete return command option Once) to focus with Remy - beer. a Riverside Studio Free Markdown Ed... Build Your Clawdbot Org Chart: From One Bot to a Full Al Team spent 9+ hours in the official OpenClaw docs and turned it into a step-by-step guide. Four multi-agent structures — from a single bot to a full company org chart with boss agents, specialist workers, and cross-team communication. Years alto recovs muy Thursday newsletter. The Al workflows, tools, and playbooks to stay loan and move fas beautiful landing page delete command option Once to focus Home /X win kemy - Door Riverside Studio with Remy - beet a Bulding a head of c Free Markdown Ed.... beautiful landing page lock delete option return command command option Once ye to focus with Remy - been X Riverside Studio Home /X with Remy - beet x Free Markdown Ed. to a full company org chart with boss agents, specialist workers, and cross-team communication. Enter your erail Yout also receive my Thursday newsletter. The Al workflows, tools, and playbooks ruse to stay ican and move fasti to put their email in lock delete option command command option What de I tel to your nên toggle the webshock late Once you've dane thi to focus or unfocus Cl Carmitoot Gra Chal Free Markdown fil to a ful company org chart with boss agents, specialint workers, and cross-team communication their email in delete return command option What do I Have to do in N807 to your nan destear oper the Openciew Pleys toggle the active switch off then back on. That fontes n wallook istanar. - the Api activation cet the statue bod you've done thet, the production ARt. to foeue or unfesue Claude to download it command Send a message send: message Execute worktiow and then an with Remy - b: X Home /x Riverside Studio Free Markdown Ed... Publish Playbook Email Delivery Editor Send a message send. messag Execute workflow delete the crazy part is... that takes me minutes to build you in on the secret all thanks to claude code alright start the timer to fee The OpenClaw Play.... Guide 1: How OpenClaw Ac Mottoshi — Al Consult Gaskell's Space...|... Digital Products / W The OpenClaw Playbook The OpenClaw Playbook Everything from your first install to a fully autonomous Al team. Who This Is For You've heard about OpenClaw. Maybe you've already tried to set it up and got stuck. Maybe your is running but feels like a worse ChatGPT. Maybe you saw someone's multi-agent setup on Twitter and thought "I want that." This guide series takes you from "what even is this?" to a fully configured Al assistant that remembers, learns, works on its own, and follows your specific business processes. Copied link to clipboard How to Use These Guide we've got lock to focus or un Guide 1: How OpenClaw Ac Edited 2d ago Digital Products / W The OpenClaw Playbook Gaskelr's Space ... Who This Is For You've heard about OpenClaw. Maybe you've already tried to set it up and got stuck. Maybe your is running but feels like a worse ChatGPT. Maybe you saw someone's multi-agent setup on Twitter and thought "I want that." This guide series takes you from "what even is this?" to a fully configured Al assistant that remembers, learns, works on its own, and follows your specific business processes. How to Use These Guides in order Esch guide Belle on the last stare at unde 1 and work theseen them seaberting Bbles 3 and a are parattet paths. Glide 2 helps you decide where to host your agent - Mac Mint ar Vos. nick the matching setth guide and skip the other. Both ontha comparge it Guide.6 the digital product lock command Optic The Openlaw Playbook built out in notion remarter net you've lied if baing to locus or unfocus Claude Playbook at/tonomous Al team. you've already tried to set 1l up and got stuck. Maybe your Maybe you saw someoner's muth-ageht setup on and we're just Tired of repeating yourself? Tell Claude to remember what you've told it using to focus or unfocus Claude Bypass permiscion. eader Optimized 8 Finish Setup *Claude Code C Share Publish • Customize site styling Search engine indexing Duplicate as template Manage all sites and links Embed this page Share via social utonomous Al team. View site ou've already tried to set it up and got stuck. Maybe your Maybe you saw someone's multi-agent setup on It even is this?" to a fully configured Al assistant that and follows your specific business processes. gonna go up here delete return shift Tired of repeating yourself? Tell Claude to remember what you've told it using to focus or unfocus Claude Bypass permission Sernan Reader Optimized & Finish Setup * Claude Code 0 Share Edited 2d ago Sharo Publish • Plus Customize site styling Search engine indexing Duplicate as template Manage all sites and links Playbook Embed this page Share via social a fully autonomous Al team. View site Maybe you've already tried to set it up and got stuck. Maybe your arse ChatGPT. Maybe you saw someone's multi-agent setup on in "what even is this?" to a fully configured Al assistant that and follows your specific business processes. and grab the delete return shift Tired of repeating yourself? Tell Claude to remember what you've told it using to focus or unfocus Claude Bypass permistions Claude Code 0 Screen Reader Optimized 88 Finish Setup we Shares Share Publish • Customize site styling Search engine indexing Duplicate as template Manage all sites and links Playbook Embed this page Share via social stail to a fully autonomous Al team. View site Glew. Maybe you've already tried to set it up and got stuck. Maybe your a worse ChatGPT. Maybe you saw someone's multi-agent setup on that." you from "what even is this?" to a fully configured Al assistant that on its own, and follows your specific business processes. published share link delete return Clicade Code Tired of repeating yoursell? Tell Claude to remember what you've told it using Bypass permissiong Copy Claude Code Finish Setup Screen Reader Optimized Share Edited 2d ago Paste in our to set it up and got stuck. Maybe your someone's multi-agent setup on configured Al assistant that business processes. Queue another message... Bypass permissions *Claude Code Finish Setup Screen Reader Optimized Share Edited 2d ago and send it off your to understand Open Chat Notion [API-get-black Show All Commands l'object": "ist, rea Start Debugging Read Index,htal Read index.htmt Read vercel. /son Read subscribe. js Thought for 2s Queue another message Screen Rea Olde 1. How OpecClam Al The OpenCiaw Playbook Dietal Produnte Edited 2d ago ShameThe OpenClaw Playbook Everything from your first install to a fully autonomous Al team. Who This Is For You've heard about Openlaw. Maybe you've already tried to set it up and got stuck. Maybe your is ruming but feels like a worse ChatGPT. Maybe you saw someone's multi-agent setup on Twitter and thought "1 want that" This guide series takes you from "what even is this?" to a fuly contigured Al assistant that remembers, learns, works on its own, and tollows your specific business processes. How to Use These Guides like what delete caps lock return control option command shift command option Access A The OpenClaw Playbook: From First Install to a Full Al Team step-by-step guides that take you from "what tven s this?" to a tully autonomous Al assistant that remembers, inams, and tostows your business processes. so it's gone From First Instal: to a Full Al Team landing page Open Chat Show All Commands to File Buiding a hiend of contente x Leus Chute Cone | Al mit The OpenClaw Playbook: From First Install to a Full Al Team take you trom what even la thie?* to a and it's even set up Open Chat Show All Commands to File winemy Jenniesetupvid X& Building a head of content o/ Learn Claude Code | Al wit From First Install to a Full / take you from "what ever the domain correctly Sper Give Chow At Conne to fie The OpeniClaw Playbook From First Install to a Full Al Tear on the web withremy_lenniesetupid. x Building a head of content or Learn Claude Code | Al with i Home /x No Free Markdown Ed... The OpenClaw Playbook: From First Install to a Full Al Team step-by-step guides that take you from "what even is this?" to a fully autonomous Al assistant that remembers, learns, and follows your business processes. and when you Bulding a head of content on x ® Alwith Remy - beehiv Learn Claude Code | Al with The OpenClaw Playbook i Free Markdown Ed. The OpenClaw Playbook: From First Install to a Full Al Team step-by-step guides that take you from "what even is this?" to a fully autonomous Al assistant that remembers, learns, and follows your business processes. You'll also receive my Thursday newsletter. The Al workflows, tools, and playbooks to stay lean and move fast enter your email Open Chat Show All Commands Go.to File Landing Access par Access page to focus or un Personal Lead Magnets and if we go into n8n caps lack Learn Claude Code | AI W Building a head of conten x Aiwithremy_lenniesetup X Personal Workflows, credentials and data tables owned by you Data tables Personal Chat beta Lead Magnets Workflows | Last updated 1 month ago | Created 1 month ago Content Playbook Email Delivery Last updated 5 minutes ago | Created 3 Marci Daily Melbourne Weather Email Last updated 2 weeks ago Created 4 March Playbook - Delivery Email Tact madator A wanks n0o. Created 19 February Help you can see Home /X Learn Claude Code | Al v Aiwithremy_lenniesetupv x/ Building a head of conten X My Free A Personal Workflows, credentials and data tables owned by you Data tables Lead Magnets Workflows | Last updated 1 month ago | Created 1 month ago Playbook Email Delivery Daily Melbourne Weather Email last undated 2 weeks ago Created 4 March Playbook - Delivery Email Building a head of conten X Editor Landing! Access page Ental Notion L Access page no watihook reg to focus or unfor Your Payback Is Ready Building a head of conter x Learn Claude Code 1 Alv Home / X Free Markdown Ed a D and that's a Claude skill the way and I've got built out for all sorts of repeatable processes like this and they literally hours a week make some skills right now

### Spoken transcript So I've just done an entire e-commerce product shoot from with inside Claude Code. Look at these photos. And this coffee table as well. Let me show you how I've done this because it's pretty cool. First things first, Claude Code cannot generate images natively. So we needed to give it a way to generate images. And I did that through this nano banana MCP server. Then I pasted 10 images showcasing the photography style I'm going for into Google Gemini. I asked it to analyze the photography style. I used Gemini because it's the best at understanding images. refined photography style prompt. Then I went into cloud code and I created this product shoot skill. I uploaded a folder of reference images for old products. Then I run the skill, say the name of the product, and it gets to work using the nano banana MCP to create all the images. Taking a boring iPhone photo of your product into something beautiful like this. If you're a founder who wants to learn and stay up to date with AI, the best place for you to be is in my free weekly newsletter. The link is in my bio. ### On-screen text TITLE: Claude Code Enter a prompt for Gemini claude code + nanobanana = unlimited ecom content entire ecommerce show you how l've done this because it's pretty cool first things first amine tone Claude code cannot fastmcp. son feat: feat: ac Code Update MO feat: add Nar claude assets Input output Open Chat Show All Commands Start Debugging generate images natively so we needed to a Type [Z) to sarch Fert s1 nanobanana-mcp-server now About mater - R reade O sTee No description, website, or topics provided star) watching BASE-URL support for custom A01 anto routing:] give it a way Defining Architecturai Photography Style attached no reterence mages Your task is to analyze the photography style only - not the objects, not the scene content. Focus strictly on: Lighting style (direction, softness, diffusion, contrast) grading and tone (warmitool balance, saturation whighlights treatment) pictoristies (ens type, focal length feet racies (framing, negative space. Write one choe thinking Enter a prompt for Genr is Tools " then I pasted 10 images How Claude Works - lavanche-Google Drive Defining Architectural Photography Style Gemini attached 10 reference images. Your task is to analyze the photography style only — not the objects, not the scene content. Focus strictly on: the ch. Lighting style (direction, softness, diffusion, contrast) Color grading and tone (warmicool balance, saturation shadows/highlights treatment) Exposure and dynamic range Camera characteristics (lens type, focal length feel aphy Sdepth of field, sharpness) Composition tendencies (framing, negative space. cropping style) Texture rendering and surface treatment Post-processing/editing style desenbe what is in the scene it dense paragraph that clearly defines Enter a p then I pasted 10 images How Claude Works- x@ lavanche - Google Drive X @ tables-cettee -GoodeD x Gemini Your task is to analyze the photography style only — not the objects, not the scene content. Focus strictly on: Lighting style (direction, softness, diffusion, contrast) Color grading and tone (warm/cool balance, saturation shadows/highlights treatment) Exposure and dy range type, focal length feel, Camera characte depth of field, shar negative space. Composition tende cropping style) Texture rendering and s deseribe what is in t Write one short, dense paragra Show thinking v Enter a promot for Gemol photography style Upgrade Gemini run attached l0 foerence nate Your task is to analyze the photography style only - not the objects, not the scene content. Focus strictly on: Lighting style (direction, softness, diffusion, contrast) Color grading and tone (warmitool balance, saturation, shadow@/highlights treatment) dynamic range geristics (lens type, focal length feel (framing, negative space. lace treatment that clearly detines. Write one shor Show thinking ~ photography style Free Markdown Ed. Productivty Dashboard| Get Defining Architectural Photography Style Gemini attached 10 reference images Your task is to analyze the photography style only - not the objects, not the scene content. Focus strictly on Lighting style (direction, softness, diffusion, contrast) Color grading and tone (warm/cool balance, saturation, shadows/highlights treatment) Exposure and dynamic range Camera characteristics (ens type, focal length feel lography S.. depth of field, sharpness) Composition tendencies (framing, negative space. desk blugg cropping style) Texture rendering and surface treatment Luxury Marb... Post-processing/editing style describe what is in the scene Mishap White one short, dense paragraph that clearly defines Show thinking Enter a prompt for Gemuni cases Online going for Free Markdown Ed. Defining Architectural Photography Style Gemini Show there V The previous iteration over-corrected by emphasizing deep tones too heavily, leading to a aar saturated look. The goal is to capture the high-contrast interplay between vibramt, motte stone and bright, refined architectiral neutrais. the pay from inape & inso the ch Refined Photography Style Prompt High-ood architectural editorial style defined by luminous, directional window lighting that maintains a bright, high-key baseline wale casting soft, sculptural shadows. The colo palette centers on a sophisticated base of warm neutrals (bone. cream and soft pege) Detring Architectural Phetorchy s contrasted ugainst localized pops of high-chroma natural atone, such an deep burgundy. rald-veined marble. or violet-hued breecia. Expostre is calibcated for a wide dynamic preserving bright. ary highlights on wall while maintaning deep saturation and rich otrass in the sonert mineral grain. The lens feel ic is 50mre to 8Somm prime with clarity on toreground lectures and a gentle roomy fall off in the barkground filten a warm. filmic white balance and a refined inate had mohinto serate a new set of images using thie talancest prompt to unaune he are correct? used Gemini lvinche- Google Drive x@ bois-cotlee-Googlet x Productivity Albaba Manutactu. me Free Markdown Ed. Defining Architectural Photography Style Gemini Show thinking M The previous iteration over-corrected by emphasizing deep tones too heavily, leading to a dark. saturated look. The goal is to capture the high-contrast interplay between vibrant, exotic stone and bright, refined architectural neutrals. the pay from inage 8 ums the eh. Refined Photography Style Prompt ok. sol am boldog house n e-com High-end architectural editorial style defined by luminous, directional window lighting that maintuins a bright, high-key baseline while casting soft, sculptural shadows. The color palette centers on a sophisticated base of warm neutrals (bone, cream, and soft grige) contrasted against localized pops of high-chroma natural stone, such as deep burgundy. emerald-veined marble, or violet-hued breccia. Exposure is calbrated for a wide dynamic range, preserving bright, airy highlights on walls while maintaining deep saturation and rich micro-contrast in the stone's mineral grain. The lens feel is a 50mm to 85mm prime with tack sharp focal clarty on foreground textures and a gentie, cresy tall-aff in the backgroued Post-processing utilizes a warm. filmic white balance and a refined matte finish, empharzing tactile surface detals and specular highlights without nuccumbing to heavy monochrome saturation or excessive shadow lilting Would you like me to generate a now set of images using this balanced prompt to ensure the lighting and color ration are correct? used Gemini Defiling Architectural Phutography $04 The provous iteration over-connected by empl leading to a dark saturated look. The goal is to catten the highvibrant, exoti tone and brighe, refined archiemural ne Refined Photography Style Prompt maintains a bright, high-key bare! the lighing and al because it's the best a 21 How Claude Works Free Markdown Ed. E Productivity Dashboard| Get Defining Architectural Photography Style Show thinking v The previous iteration over-corrected by emphasizing deep tones too heavily, leading to a dark saturated look. The goal is to capture the high-contrast interplay between vibrant, exotic stone and bright, refined architectural neutrals. Refined Photography Style Prompt High-end architectural editorial style defined by luminous, directional window lighting that maintains a bright, high-key baseline while casting soft, sculptural shadows. The color palette centers on a sophisticated base of warm neutrals (bone, cream, and soft greige) contrasted against localized pops of high-chroma natural stone, such as deep burgundy. emerald-veined marble, or violet-hued breccia. Exposure is calibrated for a wide dynamic range, preserving bright, airy highlights on walls while maintaining deep saturation and rich micro-contrast in the stone's mineral grain. The lens feel is a 50mm to 85mm prime with tack-sharp focal clarity on foreground textures and a gentle, creamy fall-off in the background Post-processing utilizes a warm, filmic white balance and a refined matte finish, emphasizing tactile surface details and specular highlights without succumbing to heavy monochrome saturation or excessive shadow lifting Would you like me to generate a new set of images using this balanced prompt to ensure the lighting and color ratios are correct? photography style prompt can you let in my clude hen and try and the out the act pest t tor my rectarans CP! Is hecated it a (Pyton pactapel, not from a Gods repo directy. The pectage a This is a PyRt package, so the Hour would be wherer the Buer how &. Let o check. •Perplexity (perplexity_search) nares The ond you/re wang l That matches you torig macty - headed a ma Would you like me to generate a new set of images using this balan the lighting and color ratios are correct? Thinking Pest Comedens v then I'went into code 1 R claude caps lock option commend commend Past Conversations v Claude Cod something worn coolone then I went into claude lock command command Enter a prompt for Genin #Chade Code dele went into option command command Ready to code? Let's write something worth deploying. Slash Commands Bypass permissions Whole Image 2000x2000 Screen Reader Optimized 83 Finish Setup 3.47MB * Claude Code producț shoot skill dele a Design Protection ata nu Italic Serif Font Recommenda... Settings & help a aime.png i ana.png Luploaded a folder help a Pe punta png of reference images dang x Bash Bath for all the products aldo.png x Thinking retry the Bash Extract o to focu before at for all the products it clanty on foreground textures and a gentle, creamy fall-off in the background. utilizes a warm, filmic white balance and a refined matte finish, emphasizing details and specular highlights without succumbing to heavy monochrome excessive shadow lifting. like me to generate a new set of images using this balanced prompt to ensure and color ratios are correct? a prompt for Gemini Thinking v our Remy Gaskell chats aren't used to improve our models. Gemini is Al and can make mistakes. Your privacy & Gemini /tiba-product-shoot v Thinking ~ start by loading all the product data and reference files needed for the Morphe shoot. Read product-catalog. j son Read dependency-chain.md Thinking Now let me get the product description for complements/morphe, check existing files, and load the prompt Agent: Find morphe description Read the file at/Users/remygaskel/Desktop/Workspaces/tiba-stone/serafini-references/_metada Bash List editorial reference images tool call errorede/too_use_errors then trun the skillLune_errorasibling tool Thinking retry the failed calls and find the product description. Bash Extract complements/morphe description to focus or unfocus Claude Screen Reader Optimized 83 Finish Setup Sarun mipt for Gemini /tiba-product-shoot v nase ta: w12 Cataiogue Hero Update Todos Generate #12 Catalogue A+B variants User picks #12 Catalogue winner Generate #01 Editorial A+B variants User picks #01 Editorial winner Generate Phase 2: #11 Technical A+B Generate Phase 2: #02 Human A+B Generate Phase 2: #03 Veining A+B Generate Phase 2: #04 Edge A+B Generate Phase 2: #05 Surface A+B Generate Phase 2: #06 Joint A+B Cleanup and verify final 8 images "Genin! 3 Pro I ("mode": "generate", "return_full_ inage": false, "node _tier": "pro", "node L Thinking Catalogue A generated. Now variant B with the same prompt. ("node": "generate", "return_fulInage":false, "node tier": "pro", "nodenane"; "Gemin 3 Pro I Both catalogue variants generated. Let me show you both side by side. Update Todos Genem the Cat User picks #12 Catalogue winner Generate #01 Editorial A+B variants User picks #01 Editorial winner Generate Phase 2: #11 Technical A+B Generate. Phase 2: /02 Human A+B. to focus or unfocus Claude Screen Reader Optimized & Finish Setup 347MB C before edits Whole Image 2000x2000 and it gets to work /tiba-product-shoot v Pnase ta: wiz Catarogue Hero Update Todos Generate #12 Catalogue A+B variants User picks #12 Catalogue winner Generate #01 Editorial A+B variants User picks #01 Editorial winner Generate Phase 2: #11 Technical A+B Generate Phase 2: #02 Human A+B Generate Phase 2: #03 Veining A+B Generate Phase 2: #04 Edge A+B Generate Phase 2: #05 Surface A+B Generate Phase 2: #06 Joint A+B Cleanup and verify final 8 images ("mode": "generate", "return_full_inage": false, "nodel_tier": "pro", "nodenane"; "Gemind 3 Pro I Thinking Catalogue A generated. Now variant B with the same prompt. ("mode": "generate", "return_ful Image": false, "modetier"; "pro", "node nane": "Genins 3 Pro 1 Both catalogue variants generated. Let me show you both side by side. Update Todos Generate #12-Catalogue-A+B-variants User picks #12 Catalogue winner Generate #01 Editorial A+B variants User picks #01 Editorial winner Generate Phase 2: #11 Technical A+B Generate Phase 2: #02 Human A+B to focus or unfocus Claude Finish Setup 347M8 & Claude Code Screen Reader Optimized before edits Whole image delete a protot for Gemini Thinking v 4 nase ta: wiz catalogue nero Update Tedos Generate #12 Catalogue A+B variants User picks #12 Catalogue winner Generate #01 Editorial A+8 variants User picks #01 Editorial winner Generate Phase 2: #11 Technical A+B Generate Phase 2: 802 Human A+B Generate Phase 2: 803 Veining A+B Generate Phase 2: #04 Edge A+B Generate Phase 2: $05 Surface A+B Generate Phase 2: 806 Joint A+B Cleanup and verify final 8 images bode": "generate", "return_ful1_Inage": false, "node itier":"pro", "node name', "Genins 3 Pro 1 Thinking " Catalogue A generated. Now variant B with the same prompt. **:"Gesial 3 Pro 1 Both catalogue varíants generated. Let me show you both side by side. Update Todos User picka #12 Catalogue winner Generato #01 Editorial A+8 variants Uner piche #01 Editorial winner rate Phese 2: #11 Technicel A+8 to focus or unfocus Claude before edits Finish Setup to create all the images delete Thinking v /tiba-product-shoot v pnose ta: Wiz Vataiogue Hero Update Todos Generate #12 Catalogue A+B variants User picks #12 Catalogue winner Generate #01 Editorial A+B variants User picks #01 Editorial winner Generate Phase 2: #11 Technical A+B Generate Phase 2: #02 Human A+B Generate Phase 2: #03 Veining A+B Generate Phase 2: #04 Edge A+B Generate Phase 2: #06 Surface A+B Generate Phase 2: #06 Joint A+B Cleanup and verify final 8 images "Goint 3 Pro I ("sode" "generate", "return fulL Image" false, "nodel, tier" "pro" Thinking Catalogue A generated. Now variant B with the same prompt. ("node": "generate", "return full_Inage"s false, "nodetler" "pro", "eode_nane": "Gemin 3 Pro 1 Both catalogue variants generated. Let me show you both side by side. Update Todos User picks #12 Catalogue winner Generate #01 Editorial A+B variants User picks #01 Editorial winnor Generate Phase 2: 811 Technical A+B Genacate. Phase 2:1t02 Human At& to focus or unfocus Claude before edits Finish Setup: 347MB Cla to create all the images delete Bash Rena Now final cle Update Tode Generate Generate User picke Generate: Generate i User picks Cleanup an Bash Final clea Exit code lata Bash Final clean Update Todos to focus or un before edit taking a boring iPhone photo Whole aidang X Bash Rename wit Now final cleanup a Update Todos Generate POt fe User pick User-picks 102 He Cleanup and verify Bash Final cleanup an Exit code 1 (eval): 2: no mat Bash Final cleanup and thumb. Jpe (eval)ils no sate scene drer-ar-ag 10 re Update Todos taking a boring iPhone photo to focus or unfocus ( Whole Image aldang x Bash Rename w Now final cleanup Update Todos Generate #12Generate #01E Generate Phese Regenerate #06 Regenerate 402 Cleanup and veri Bash Final cleanup a Exit code 1 (eval):2: no a Bash Final cleanup anc _thumb. Ji (eval)it: no mat dnar-ar-@ 10 m Update Todos to focus or unfocus of your product Whole Image aidang X /tiba-product-shoot v Bash Rename winner, dele Now final cleanup and verific Update Todos Generate #12-CetalogueCleanup and verify final 8 ir Bash Final cleanup and list all fi _thumb-jpeg _a-l Exit code 1 (eval):1: no matches fou Bash Final cleanup and verify bene (eval)ils no matches foun Update Todos to focus or unfocus Claude into something before edits Whole Image /tiba-product-shoot v Bash Rename winner, delet Now final cleanup and verific Update Todos Generate #12-Getalogue Generate Phose-2:411-Tec Generate Phase-2-#03-40 Cleanup and verify final 8 in Bash Final cleanup and list all fi Exit code 1 (eval):1: no matches four Bash Final cleanup and verify (eval):2: no matches found Update Todos to focus or unfocus Claude before edits into something /tiba-product-shoot v Bash Rename winner, delete los Now final cleanup and verificatior Update Todos ping Generate #01-Editoriet-A+B-vel Generate Phase 2-#11 Technic Regenerate #02-Human (seater Cleanup and verify final 8 image Bash Final cleanup and list all files Exit code 2 (eval)il: no matches found: Bash Final cleanup and verify (eval)il: no matches found: • Update Todos to focus or unfocus Claude beautiful like this before edits Whole Image you're a founder who wants to learn and stay up to date with Al the best place for you to be is in my free weekly newsletter the link is in my bio

### Spoken transcript So unless you've been living under a rock, you've probably heard of Obsidian. No, not that kind. It's literally just the most stupidly simple note-taking app you've ever seen. But when you plug it into Claude Code, it becomes an entire neural network for your AI agent. Connects all your thoughts, projects, people, meetings, and ideas in one place, allowing Claude Code to behave in ways that I've never seen before. Like every morning, I have Claude plan my day. And last week, it came back with, you've been talking about rewriting your welcome email since January. You've mentioned it in four different notes. started, that's your priority today. Or the other day, I was building a landing page and Claude went into my vault, it found a transcript of a client call I had three months ago and pulled out an excerpt to use as a testimony on the landing page. But that's the kind of wild shit that happens when your AI actually has context. I wrote a simple guide on my Claude Obsidian setup and you can even give it to Claude and say, set this up for me. Comment the word Minecraft and I'll send it to you. ### On-screen text Obsidian + Claude Code is insane living under a rock that kind literally just "So unless your ve been living under a rock, you've probatty heard of Obsidian a stupid sumple note-taking app - ike Apple Notes Our vihen you plug at into Claude Code, It hecomes something else entrey Thes is an entire neural network for your Al agent. it cortects all your thoughts - broncts. Aroving Claude Code to behave in ways vo revor seem bofore. every motting have Clapde plan my day last week it came back with - You've been talking about retting your welccene emai since Jensary. You ve mentioned it in four different notes. You haven't afarted. That's your prionty today: the other day I was bullding a landing page and Clude went into my vault, found a transcripe from a clent call these months ago, and pulled out a sentence to use as a testimortal on the page id completely forgotten that call ever happened That's what happers when your Al actually has context. I wrote a simple quide on my claude cores ohud an setup, you cart even give it to claude code and ask it to set this up for you, Comment obsidian and il send it over the most stupidly simple Ai with Permy So unless you ve been le a studid simple note-ta when you pog to nto Ca This is an entire neural nets Blowing Claude Dods to ser Like every morning name Caus And la came bac cater day I wat call thre stupidly simple Obsidian _System with Remy Daily logs Meeting Notes "So This is. people, Like every And last we since Januar priority today the other di from a client ca page. I'd compl That's what happ code x obsidian s Comment obsidian note taking app Obsidian _System with Remy Daily logs Meeting Notes *So unless yor a stupid sim when you pl This is an entire n people, meetings Allowing Claude Ca Like every morning 1 And last week it came since January. You ve priority today! the other day i was b from a client call three m page. I'd completely forg That's what happens whet code x obsidian setup, you Comment obsidian and ill se note taking app "So unless you ve been ling under a rock, your ve procably heart of Condie when you plug it into Claude Code, becomes someng ese emire This is an entire neural network for your Al agent. it connects all your trougs - projec people, meetings - together. Allowing Claude Code to behave in ways rve never seen before Like every morning | have Claude plan my day. And last week it came back with - You ve been taking about rearting your wecome eme since January, You've mentioned it in four decent notes. You tavent sared that s you priority today: the other day i was building a landing pege and Claude wart into my vaut, found a transcre from a client call three months ago, and duled out a sentence to use as a page. I'd completely forgotten that cal even happene That's what happens when your Al actually has contest. wrote a s code x obsidian setup, you can even give it to claude code and ask te Comment obsidian and & send it over. gude on my cauti you've ever seen when you plug it into Claude Code it becomes entire neural network success And ED B. 9 Graph wen neural network Project Succession And ED B... 3º Graph view Model Extrapolation Graph view for your Al agent connects all your people and ideas in one place allowing Claude Code to behave in ways that I've never seen before like every morning have Claude plan my day and last week it came back with you've been talking about rewriting your welcome email since January you've mentioned it in four different notes you haven't started that's your priority today the other day was building a landing page and Claude went into my vault it found a transcript of a client call l had months ago and pulled out excerpt to use as a testimonial on the landing page that's the kind of when your Al actually has context wrote a simple guide Claude x Obsidian setup and you give it to Claude set this up for me comment the word and I'll send it to you

### Spoken transcript Hey guys so in today's video we're covering the different ways to actually make money from home using AI in 2026. There are thousands of ways to make money with AI right now but I've sorted through them all and come up with my own list based on my own experience of building AI businesses for the past three years and what's actually working for real people inside my communities. Everything we're talking about is tried and true and I'm going to be showing you the real tools, the real platforms and exactly how you can do each one of these yourself. Honestly you need to know what the latest stuff is right now because there are so many new models and tools coming out that the opportunities to look completely different than what they did even six months ago. So let's get into number one. So the first method is what I call the instant website flipper. Here's the idea. Your local dentist, local plumber, local accountant, their website probably looks like a fossil from 2005. They know it, you know it, the customers know it. And they also know that a web design agency is going to charge them $10,000 and take three months just to fix it. But you can now do it with AI in just about two minutes. Okay, so to start off with flipping these websites, you're going to want to head to Aura.build. No affiliation with this tool, but it's a really, really good platform creating beautiful websites with AI now to show you an example I've found some random dentist business near me in Auckland so the way to do this is you need to take the information off the website somehow you can either kind of copy everything or the better way is to actually right click and go to view page source we're going to copy all of this page source we're going to head back to Aura here and then we're going to go to the template section and here we actually have a dental template that we can use so we can click on this take a look at it looks great I'm going to go use for free here then I'm going to template. Then we just need a quick prompt saying that we want to redo the website. Here's the source code and here's some tips and tricks on what we want you to do with it. So remember that I pasted in all of the source code that we got from the website over here. So I've added that into the prompt and now it's going to rebuild us the website using more than information, but in that template style. And then just like that, we have a website that looks far better than the previous one. It's even got all the pictures of their team here from the website. It's got the services they have, their assessment and their logo as well. And with the old one and the new one side by side here, I think you can agree just how much better and more modern this looks in just a few minutes so now you've got this beautiful finished website you're gonna walk into their office and then you're gonna show it to them on their laptop and say this is your new website you can start selling these at $500 get a bit better charge a thousand or two thousand you do two or three of these a week and you're already hitting ten thousand dollars a month and from there you can upsell them on other stuff like the other methods that I'm about to share with you in this video alright let's get into number two method two is my personal favorite and it's probably the easiest way to get started if you're not technical at all you don't know how to code we've never had any exposure to that sort of stuff before it's called an AI tools audit. And here is the problem that you're solving. Every small business owner right now knows that they need to be using AI. They're seeing the headlines, their competitors talking about it, but they have absolutely no idea what tools to use because there's thousands of AI tools out there and they don't have the time to research them all and figure out which one works for them. They're paralyzed, but they know they need to do something and you can become the person who tells them exactly what to do when it comes to AI tools. So the way this works is simple. You sit down with a business owner, you ask them about their biggest bottlenecks, their most annoying tasks and the stuff that really eats up a lot of their time and their team's time then you go away and you find three to five ai tools that solve those exact problems you put it into a simple report for them you hand it over and you get paid that is it you're not building anything and you're not coding anything either you're just selling ai clarity to these small businesses i've got this tool here which is called there's an ai for that which is going to give you everything you need as a complete beginner to find tools that solve problems for businesses so say for example you do an audit and they say hey we've got a real problem with invoicing we're spending so much time in our finance department tracking down and bouncing around and trying to get the right invoices so that we can reconcile our transactions. So you can come in here and click search and go invoices. And here we have the invoices task. We can scroll down and there's all of these different AI tools that we can use or try out or investigate for them to see if it's a good fit for their workflow. So here's one called get invoice. We can look at that AI powered invoice extraction 10 times faster. We can click on their website here, check it out. AI accounting automation that saves you hundreds of hours. So just by looking at it, it looks like you can add in all of your Gmail accounts go and do the emailing for you to get the invoices sent in order to reconcile those transactions and that's just one simple example if you go to tasks here you can see all of the different categories productivity software text images marketing you literally just need to know how to search and how to put together a simple report then you can start charging two hundred dollars five hundred dollars for these just starting out it takes a couple of days for you to deliver when you're just starting out but as you get better you can raise your prices to a thousand dollars two thousand dollars five thousand dollars at my agency morningside ai we've done these kinds of ai audits for $60,000 for a 200 person company. And here's the thing, the data from my community shows that the average time to land your first client with this method is just nine days when you're using the right type of outreach method to get those clients. And because I think this is such an incredible opportunity for beginners to get into AI in 2026, I've just released a full free course on exactly how to do one of these audits. The outreach templates to get your client in nine days, the audit structure, the video that explains this model and how to get into it will be in the first link of the description, but let's keep moving. All right, so method three is where things get really fun. need visuals they need product photos and social content and video ads ugc style clips they need this stuff constantly and traditionally this stuff is very expensive and slow for them to get a proper agency will cost thousands of dollars and will take weeks to organize you need to get cameras and actors and locations a whole crew just to get these photos of products and now one person with a laptop and the right set of prompts can do all of it and i'm talking product photos that look like they came from a professional shoot ugc style ads without any real actors and cinematic shots without a camera my go-to platform the moment for me and my content team is in video who are also the sponsor of this video so shout out to the team there but i'm more than happy to be showing you guys the exact tool that i'm using in my creative team right now so video you can get there the link will be in the description to sign up and you can either sign up to create an account or you can log in if you already have one and once inside you'll see this input where you can add in a prompt this will trigger in videos video agent which is going to make a bunch of creative decisions for us under the hood and is capable of generating complex longer videos for us for now since we want to keep things short and sweet let's head to the agent and models tab so i came on here and just grabbed a mug I took a photo of my mug and then I put it into in video and I was able to generate all these different types of content in just a few minutes starting with some product photos that would be suitable for e-commerce you can see you can put them with literally just a simple prompt like this you can get crazy kind of Black Friday ads like this using some of their trending formats which you can do by just coming into here clicking on trends I also did some really luxury style ads like this just a bit of footage and you can even generate UGC clips like this hot. I use it every single day. And these are obviously quite quick and dirty, but it shows you in just a few minutes what I've been able to generate with just a few very basic prompts. So if you make this your platform and you really spend time learning how to use all the features, learn which models are best for what, you can literally become a one-person AI creative studio using a tool like this. And if you're serious about it, you can invest in a generative account, which is $100 a month. So generating those literally took me a few minutes, but you can see how if you really lean into and learn a tool like NVIDIA, you can become essentially a one-person creative studio, and have a handful of clients that you help with those kind of creative needs. To get this off the ground, you can approach small businesses and offer them a simple monthly package, something like $2,000 a month, which will get them 150 images and 10 videos. And you're basically just going to be on standby. They're going to send their requests and you'll deliver them the creatives that they need fast. Once you stack a few of these clients, you're at 10K per month pretty quickly. All right, getting into method number four. So method number four is one of the fastest growing opportunities that I'm seeing right now. And it's writing the enormous AI agent's weight. you've called the business and nobody picked up or you got put on hold forever or you just had a simple question and couldn't get an answer and this kind of stuff happens constantly and every time it does that business is losing money because you're going to end up going to their competitor you can solve this by building them an ai phone line that answers customer questions 24 7. we're going to go on to my go-to platform for this at the moment which is 11 labs you can log in they do have a free plan literally two minutes we're going to be able to build an ai voice agent that does customer support for our friends at parnell dentistry again no with them they're just a target that we're going for this time so a new agent here we can go to a business agent healthcare and medical then we're going to make it a customer support we're going to call this par now dentistry support agent we can put in par now dentistry pop them in there right so if you want to set this up super quick you just need to put a really good prompt in here just to give it an idea of the main goal so you're a customer support representative for par now dentistry you answer questions about the company's service location team etc knowledge base you have been provided with because we're going to give it a knowledge first in a sec if people ask about booking or pricing you redirect them to the website at the link here so we can create the agent 11 labs going to do most of the work for us thankfully and there we go 11 labs has written out our prompt for us here now we can just go to knowledge base now off camera i just went on to partner to introduce websites i copied a bunch of the pages and then i pasted into chat gpt asked it to write me a knowledge base document with all the key information that an agent would need to know then i can just copy this add a document create text paste it in there then we can go back to our agent make sure that the prompt is mentioning the knowledge base that you have access to a knowledge base about pile dentistry your primary goal is to answer questions if customers inquire about bookings or pricing send them to here and we can add a nice conversational first message here then we can configure the voice i'm going to try to find a new zealand we've got our john here great llm provider we can just keep it as gemini 2.5 flash and then we're ready to give this a spin Hi, welcome to Parnell Dentistry. This is Mike speaking. How can I help today? Hey Mike, I was just wondering what your mission is at Parnell Dentistry? Hello. Our mission at Parnell Dentistry is to prioritise our patients' oral health with compassion. We aim to deliver a pleasant and stress-free journey. Okay, yeah, yeah, mate, that sounds great. Now I really want to know who's on your team. Give me one of the dentists on your team. Certainly. One of the dentists on our team is Dr. Kavendra Naidoo. Excellent. Appreciate it, mate. So we just built a working AI customer support line in about a minute. But to turn this into a business that you can make money from consistently, you need to find some local businesses, build them a quick demo using their actual information like I just did there. And then you walk in and let them talk to it. Let them ask questions about their own businesses and watch their faces as they realize what's possible. And this kind of stuff is just the starting point, really. Once you understand the basics, you can move into much more advanced AI voice agents that can book appointments over the phone, upfront and then hundreds or thousands of dollars or more per month for every single client in order to manage and improve that system for them if this stuff seems interesting to you and you want to learn more about this path into essentially learning how to build with ai that same full course on starting ai business which i mentioned before is going to be linked below that's going to explain the whole model to you so you can get into it so i highly recommend you check that out after this now method 5 might sound basic but do not sleep on it because companies are spending thousands of dollars on ai tools like chatty bt and claude and all the other ones but their teams are still using them to do basic stuff like writing birthday messages or drafting emails and it's a massive waste of these tools and these businesses are desperate for someone to be able to teach their people how to actually use and get value out of the stuff and i know firsthand that this stuff works because it's one of the things we sell at my agency morningside ai but if you're looking to get into it yourself here's the play what you need to do is first and foremost reach out to people that you already know like your old bosses your former colleagues friends who run businesses and offer to do a free workshop in exchange for a testimonial then you're going to walk in minds by taking one of their actual repetitive tasks and showing them how to do it much faster with ai right there in front of them and as you may have noticed there's a bit of a pattern here of showing rather than telling when it comes to communicating the value of your ai services so that free workshop that you do is essentially a trojan horse and once you've proven your value with it the companies inevitably want more help from you like one guy from my community mateo has made over four hundred thousand dollars with this exact model because seventy percent of the companies that he does these basic training workshops for become clients for much bigger projects later on you can start with free workshops to build testimonials then go on to selling 500 to a thousand dollar sessions then you can do half day intensives for two to five thousand dollars and you land a few recurring clients in there and you're easily at 10k per month so there you go five real ways to make money from home with ai in 2026 but if i had to recommend one for most people it would be the ai tools audit like i said it's the easiest to start you don't need any technical skills and the demand for it is just insane right now and like i said i believe in this so much that i released a full guide on youtube breaking down exactly how to start an ai business from scratch that's based on this AI tools method. Plus, I created an entire free course that comes with it to give you everything you need, like the templates, the Albreed scripts, and a full step-by-step process to deliver your first AI tools audit. So you can watch that up here. Highly recommend it, guys. Or it's going to be in the first thing in the description if you can't click there. This is the opportunity, guys. Thank you so much for watching, and I'll see you in the next one. ### On-screen text Lasury Salon Landing Page Sereetrading Chadenge Cheawy Lee Banana Pro, newest image generation tool Google Al Pro and Ultra subscribers can use the Nano Banana Pro in Search's Al Mode you can do it with AI Create beautiful designs Generate top-ter landing pages in seconds. Watch video. beach li propers Adapt tone Nature Pertermance APt Polyfliis →y Dental Clinic Landing Page Template dept free "Demaal Canic Landing Pace Home Advanced Dental Care Designed for Comfort Advanced Dental Care Designed for Comfort Experience gentle treatments, modern technology, and a How would you ike to change this page? safe, hygienic clinic environment tailored for your peace of "Change this page to features" mind. to a pricing page with pricing. Book Appointment → Explore Services Below Is the source code of Parnell Dentistry's Website, want you to take all their website copy and positioning and service Home Advanced Dental Care Designed for Comfort Advanced Dental Care Designed for Comfort Experience gentle treatments, modern technology, and a would you like to change this page? safe, hygienic clinic environment tailored for your peace of sange this page to features" this page to a pricing page with pricing. 4/body» can Crea the moderrized, professional version of the retaining the original content, images, and Pages -home v Preview Design Code Book Now Experience The Epitome Of Dental Care Parnell Dentistry, we combine advanced technology with a compassionate approach to provide you with a pleasant journey to excellent oral health. Pages - home v E About Us Book Now Michael Georgy Our Expertise Wisdom Tooth Removal Dental Sedation Expecence painiess wisdom foots removal by commona cent warney, task our some More w Experience The Epitome Of Dental Care Experience Parnen Dentistry The Epitome Of Dental Care Our Mission Our Team Our Mission Genius Ways to Make Money TECHNICAL Earl Fools try Possibilities Easily, Possibility biggest bottlenecks most annoying tasks RECOMMEND Free mode Accounting 6l Specialized tools = Invoice Cio Getinvoice s Seep nunual Invoice Typing was Al scanning Sanand Ty ago Free mode Personal Accounting 6l Specialized tools Tong * fres • from $50 No pricing We ute it at Ucademy with 10 gma Invoice Data extraction Trusted by teams aroond the world Sign up with Google Get started Free mode sign up a Marketinol Text we Software Health -Coding m INCREDIBLE OPPORTUNITY Genius Ways to Make Money product photos social Peterate my Video" nerate my Video Create short nded Ai videos Millionaire Search beautu tudo-Quality product photo of his mug in loury dormatic setangs utable for a high end ecomenance stone Cole a forus Desere wear you want to genenten video a Genius Ways to Make Money ARCHITECT Agents Platform spor) Home Crested by Cressed at = Name Knowledge Basel Support agent Thone Nombers Complete your agent Name your agent, describe its godl, and optionaly add your website Agent NamePamell Denistry Support Agent Main Goalfor any service using the 30minute appointment slot as long as there is availablty for that Agent Agent Name • Parnell Dentistry Support Agent Website (Optional) Main Goal • You are a customer supportrepresentative for parnell dentistry, you answer questions about the company's services, location, team etc from the knowledge base you have been provided with. If people ask about pricing or booking, then you redirect them to the website Chat only Audio will not be processed and only text will be used Create Agent Back Agent Name Parnell Dentistry Support Agent Website (Optional) We'll only access publicly avalable information from your website to personalize your agent. Main Goal You are a customer support representative for parnell dentistry, you answer questions about the company's services, location, team etc from the knowledge base you have been provided with. If people ask about pricing or booking, then you redirect them to the website Chat only Audio will not be processed and only text will be used Back Create Agent Complete your agent Nome your agent, describe its gool, and optionaly odd your website Pamet Dennistry Support Agent Websine (Optional) Main Goal - Parell Deetistry Support Agent/ Main Lve100% tae View new foatures > Agent Eric - Smooch, Trustworthy Primary advice er make degroses. If you ace undure about an answer, poltety state tuat you do not have the informasion and Type (C to sdd variubies a Set tesezone Fest message The frst message the egent will sty, If empty, the agent wil wat for the user to start the conversation: Disclosure Reguremsent Preview Agent Knowledge Base document No documents found And document Panel backing. New elim. You we hey. he plan kotones hout the cord as seces. iron. and hm You ownre tha user's cral beath and approach it with compassion. You are assisting customers via a coversational Al interface. Yue have access to a knowledge base about Parnel Dentiatry, including information about the team, services, and location. The user may be inquiring about general information, pricing, or booking appointments. Your responses are polte, peofessional, and isformative. You speak clearty and concisely, providing accurate itformation and helpful guidance. You Your primary goal is 8o answer cuntomer questions about Parnel Dentistry's services, Iscation, and team bassed on the provided Anowledge base. I cutonens nure so prare or cocan apponents meet sam to the wrote sinewww caneous corn provide informabon that is factal and accurabe based on the provided knowledge base. Do not provide medical advice of make dagsoses. # you are unture about an answer politely state that you do not have the informition and suggest visting the webste or contacting the clinic directly. engage in offensive, discriminatory, or inappropriate language. None Patel Acdim. Naw earn. You we renoy. he put in onto east hot the toroams serves. Hot m. ma Him you mixere th You are assisting customers via a conversational Al interface. You have access to a knowledge base about Partell Dentistry, including information about the team, services, and location. The user may be inquiring about geceral information, pricing, or booking appointments. Your responses are polibe, professional, and informative. You speak clearly and concisely, providing accurase information and helpful guidance. You Your primary goil is to answer customer questions about Parnel Dettistry's services, location, and team based on the provided knowledge base. If note Host Prong of booking appo names Tess cam to the wroste a captaiwww bank outsy cord you are unture about an answer, poldely state that you do not have the information and wagest visiting the wedde or contacting the clinic directly engage in offensive, discriminatory, or inappropriase language. Type 1 to add variabled Parnell Dentissry Support Agent / Main Live100% Od-mtvagrg: Agent Sysiem prompt ? Select te ElevenLabs volces you want to usel Eric - Smooch, Trustworthy Primary Language You are assisting customers via a conversational Al interface. You nave access to a knowedpe base about Parnet Sype (c to add vaclable Vest message The frit message the agent wit sity, If empty, the agent wil wat for the user to start the conversation, Disclosure Requrertent Cenni-s flash welcome to Pamel Dentistry this is Mike soraking How can i hel todw? Type (( to add vaniables Agent voice Parnell Dentistry Support Agent Main Live 10% 0_ahapig: Select a voice Agent Now View newfeatures > is model here Systems prompt a Eleven Turbo You are a customer sucest rearesentatie for Parnell Dens stry, a family-onensed oneral practice with over two Speed compasy's services, location, and leam. You prioritize the user's oral health and approach k with compassion type c to add variables Defaut personairy a Se message The inst message the gent witsy it empty, the agent wit wat for the user 1o start the conversaron, Disclosure Red Type ( to add variables Parnell Dentistry Support Agent. Main Lee 100% 1 trageo Ageet Wocklow Me Branches Nee Knowledge Base Aeslysis Tools Tests New Widget Security Advanced No View new features > Agent Systers prong Select the Eleveolaos voices you waot to usel the apes company s services, location, and team. You proritize the user's cral heath and approach E with compassion. Language Environment You are assisting customers via a conversational Al interface. You have access to a knowledpe base about Parnel English Delic Type If to add veciabses soational language That message The irst message the ogent will say. If empty, the agent will mut for the user to slart the conversaron, Disclosure Regurementi Type if to aad variabres Experience Switch to chat mode The Epitome Of Dental Care Parnell Dentistry Our Mission Talk to interrupt Our Team shoe shit the comes soon Parnell Dentistry Switch to chat mode Our Mission Our Team Talk to interrup: shoe to shite comers soon call in retainer income Training claude ### Video description & links 🚀 My Latest AI Business Guide: https://youtu.be/GTWWNZyIsSc 📹 Try out Invideo for image & video generation: https://invideo.io/i/LiamOttley 📚 Join the #1 community for AI entrepreneurs, get all my resources & connect with 280k+ members: https://bit.ly/4600IoH 📈 Become a Wildly Profitable AI Entrepreneur: https://bit.ly/3NluM81 🤝 Ready to transform your business with AI? Let's talk: https://bit.ly/4jNR9PF 📋 Get our FREE 14-day playbook for finding high-impact AI opportunities in any business: https://bit.ly/14-day-playbook- Five practical ways to make money from home with AI in 2026, focused on beginner-friendly AI business ideas that don’t require coding, an office, or a technical background. I’m breaking down how to use AI to upgrade local business websites, running AI tools audits for small and mid-size companies, creating AI-generated content and ads from a laptop, building AI phone agents for customer support and bookings, and getting paid to train teams on tools like ChatGPT and other AI software. These are remote AI income ideas people are using right now to earn online, land clients, and build scalable AI side hustles or full-time businesses from home. ⏱️ Timestamps: 00:00 — Making money from home with AI 00:33 — Method 1: Website flipping 02:21 — Method 2: AI tools audits 04:56 — Method 3: AI content & ads 07:21 — Method 4: AI phone agents 10:49 — Method 5: Paid AI training 12:12 — Free resources + next steps

### Spoken transcript Search blogCtrl+K BLOGAgentic Coding in 2026: A Practical Guide for Big Code May 21, 2026 Matt Tanner Agentic Coding in 2026: A Practical Guide for Big Code Learn what agentic coding is, how AI coding agents work in real engineering orgs, and how to give them the codebase context they need to ship safely. Something shifted in 2025. AI coding tools went from suggesting the next line of code to writing whole pull requests, running the unit tests, and opening the next one. The label that stuck was "agentic coding," and by mid-2026, many large engineering organizations were experimenting with at least one agentic coding workflow, whether through IDE assistants, background coding agents, PR agents, or internal tools. The trouble: techniques that work on a side project fall over the moment you aim them at a 2,000-repository monorepo with twelve years of accumulated decisions. This guide is for senior engineers and platform leaders making agentic coding work at real scale. We'll define the term, separate it from "vibe coding," walk through how the loop executes, and spend most of our time on why agents quietly break at the 80% mark and what context infrastructure stops that from happening. What Is Agentic Coding? Agentic coding is software development where an autonomous AI agent plans, writes, tests, and iterates on code with limited human intervention, using tools (a shell, a test runner, code search, and version control) to complete complex tasks across the development environment. Instead of typing every line and accepting one suggestion at a time, the developer describes an outcome, and the agent runs a loop until the outcome is reached or it gets stuck. That definition draws a line that other AI coding categories don't cross. Autocomplete predicts the next token. Chat answers a question. AI agents take actions: they read files, call tools, run commands, observe outputs, and decide what to do next. By 2025, Anthropic, OpenAI, Google, GitHub, and others had moved from autocomplete and chat toward agentic workflows that can inspect repositories, run commands, call tools, and propose pull requests. Sourcegraph's CodeScaleBench work focuses on measuring how agents perform on large-codebase and multi-repo tasks, where retrieval quality affects execution time, retries, and cost. How agentic coding differs from autocomplete and chat-based AI The simplest way to feel the difference: watch what each tool does when you say "fix the failing build." Autocomplete does nothing. It needs you to start typing. Chat-based AI reads the error you paste in and suggests a fix. You apply it manually. An agent opens a terminal, runs the failing test, reads the stack trace, greps for the broken function, edits the file, re-runs the test, and reports back when green. The behavioral leap is autonomous tool use: the agent decides what to do next based on what it just observed, the same way a human engineer does. Agentic Coding vs. Vibe Coding (and Why The Distinction Matters) "Vibe coding" was coined by Andrej Karpathy in a February 2025 tweet describing "a new kind of coding" where you "fully give in to the vibes, embrace exponentials, and forget that the code even exists." Karpathy was honest about the scope. He was talking about throwaway weekend projects. The term escaped that container immediately. By the end of 2025, it was Collins Dictionary's Word of the Year, applied to everything from prototype tinkering to production deploys. That's where the confusion starts, and why senior engineers bristle when the two terms get mashed together. Vibe coding is a posture: trust the model, don't read the diff, keep prompting until it works. Agentic coding is an architecture: the model is wired into tools, runs in a loop, and produces code that a human reviews against a definition of done. The difference is who owns correctness. Simon Willison's writing on agentic engineering frames the practice as professional engineers using coding agents to amplify existing expertise, not replace human judgment with vibes. Vibe coding works when the cost of being wrong is zero. Agentic coding has to work when the cost of being wrong is a customer outage. DimensionVibe codingAgentic coding Human roleTrusts the model, ships if it runsReviews the diff, owns correctness Code reviewOptional, often skippedMandatory, often automated as a gate Verification"Looks like it works"Tests, types, CI, sometimes a second agent Best fitThrowaway prototypes, side projectsProduction code in a maintained repo Failure modeBad code shipped without anyone noticingBad code caught at PR review or in CI If your team ships into a codebase other people maintain, you're doing agentic coding whether you call it that or not. Treating the agent's output as code that needs review, not as a finished product, is the line. How an Agentic Coding Loop Works Every agentic coding system runs a variation of the same loop: Prompt. The developer states the goal. "Add rate limiting to the checkout endpoint." Context gathering. The agent searches the codebase, reads files, and pulls in docs or tickets. Plan. The agent drafts a sequence of edits and verifications. Execute. It writes code, modifies configs, and shells out to tools. Test. It runs the test suite, linter, or type checker. Refine. Failures feed back as new observations. The loop runs again. Output. When tests pass and the plan is complete, it surfaces a diff for review. What makes one agent better than another is rarely the model. Many tools draw from the same frontier model families, so the differentiator is often the system around the model: retrieval, tool use, planning, sandboxing, and review. What varies most is how good each step of that loop is, especially step two. The role of context (codebase understanding, repo-wide search) Context gathering is where most of an agent's output quality is won or lost. A model can only reason about code it can see. If the agent searches the local working directory and finds three relevant files, it plans on three files. If the real change affects 17 files across 9 repositories, the agent doesn't know and won't ask. It produces confident, locally correct code that misses two-thirds of the work. This is the difference between approximate retrieval (embeddings, vector similarity, "files that look related") and deterministic search (exact symbol references, every callsite, every interface implementer). Approximate retrieval is fine on a small repo. On Big Code, it returns plausible-looking results that miss cross-cutting impact, and the agent ships plausible-looking code with latent bugs. Sub-agents and parallel execution Modern agents delegate. A planner decides what needs to happen, then spins up multiple agents to handle independent slices: one writes the migration, one updates unit tests, one drafts docs. Stripe's internal "Minions" workforce, described by Alistair Gray, follows the same shape: many small agents coordinated by shared context. Sub-agents amplify whatever context layer the parent has. Bad context yields parallel wrong answers faster. The 80% Problem: Where Agentic Coding Breaks Down at Scale Here is the pattern anyone running agents at scale recognizes. The agent finishes a task in five minutes, the diff looks clean, the tests pass, and you merge it. Three days later, an unrelated team's CI starts failing because a downstream service still expects the old function shape, and your agent had no idea that service existed. This is the 80% problem. AI coding agents reliably do the visible 80% of a task and miss the invisible 20% that lives outside their context window. The missed 20% is the same categories every time: auth middleware wrapping the changed function, API DTOs serialized at a different layer, audit logs recording state transitions, integration tests in a sibling repo, frontend guards mirroring backend permissions, and migration scripts that need regenerating. In the illustration above, an agent without full codebase context edits models/user.go and the store, declares done, and misses five things: auth middleware that doesn't check the new role, an API DTO that never returns it, the audit logging path, the frontend admin guard, and the invite flow that creates users without a default role. With cross-repo search spanning the full estate of repositories, the same task surfaces references across multiple layers, and the agent edits the additional files it would otherwise have missed. The agent didn't get smarter. It got more eyes. This paradox of agentic coding generating more code is borne out in practice: we found that 84% of large enterprise accounts saw a steady increase in lines of code after AI rollout, and developers responded by searching more, not less. The agent's productivity ceiling is set by what it can see, and on Big Code, what it sees by default is a fragment. The 80% problem isn't a model limitation. It's a context infrastructure problem that shows up in three ways: Cross-cutting code changes. Anything touching more than one service, repository, or layer. Hidden technical debt. Subtle overrides, custom decorators, and sibling microservices that the agent never opens. Monorepo blind spots. Searches limited to the current working directory miss the rest of the tree. If your team is hitting these and blaming the model, you're debugging the wrong layer. Practical Workflow for Engineering Teams The teams getting the most from agentic coding aren't the ones with the most expensive models. They're the ones with disciplined workflows that play to agent strengths and design around weaknesses. A useful baseline: Scope the task. A senior engineer writes the prompt, sets acceptance criteria, and decides whether the change is local or cross-cutting. Equip the agent with context. Point it at codebase search, internal docs, tickets, and architectural conventions. Run the loop. Let the agent plan, edit, test, and refine. Watch for a clean exit or a stuck state. Review the diff like a human pull request. Read it. Run it locally. Check the unit tests it wrote. Gate on CI. No agent commits land without passing the same checks that human commits have applied. Audit afterward. Track agent activity by agent, prompt, and code changes. Treat the agent like a contractor with commit rights. Steve Yegge's brute squad framing captures the new developer role in an agent-led workflow: less typing code, more steering and verifying many concurrent runs. The work doesn't disappear. It moves up a level. PhaseWithout codebase contextWith full codebase contextRisk addressed ScopingEngineer guesses blast radiusSearch shows every call site up frontUnderestimating cross-cutting impact PlanningAgent plans on visible files onlyAgent plans across all dependent layersMissed integrations ExecutionEdits one or two reposEdits all affected repos as one unitPartial rollouts TestingLocal tests passSibling-repo integration tests includedLate CI failures ReviewReviewer rebuilds context manuallyReviewer sees the same scoped graphSlow, error-prone PR review AuditHard to reconstruct what changedCode Insights tracks adoption over timeBlind compliance gaps A useful pattern: when an agent says it's done, search the codebase for any other usage of the symbols it touched. If something turns up that the agent never opened, the task isn't done. Tools and Infrastructure for Agentic Coding The 2026 agentic coding stack splits into three layers, and most teams use something from each. A category sketch, not a buyer's guide: Coding agents run the loop. Common options include Anthropic's Claude Code, OpenAI Codex, GitHub Copilot's agentic workflows, Google's Gemini CLI, Amp, and open-source autonomous agents. They differ in model defaults, execution environment, sub-agent design, and how they handle context beyond the local workspace. IDE integrations and chat assistants are AI tools that generate suggestions and answer questions in the editor. They're useful for short-horizon tasks but not autonomous in the loop sense. Context and infrastructure layers are the substrate that makes AI agents work in a large org: indexing, code search, code intelligence, repository-wide knowledge. The Model Context Protocol (MCP), introduced by Anthropic in late 2024, is an emerging interface that gives compatible AI agents a standard way to request context from tools and systems that expose MCP servers. A useful mental model: agents are the engine, MCP is the wiring, codebase context is the fuel. A great engine on bad fuel still stalls. How to Give Agentic Coding Tools the Context They Need (Featuring Sourcegraph) If the 80% problem is a context problem, solving it means giving every agent the same view of the codebase a tenured engineer has on day 1,000. This is what Sourcegraph is built for, and it plugs into whichever coding agent your team has standardized on, whether that's Claude Code, Gemini CLI, or another option. Sourcegraph indexes every repository across your code hosts (GitHub, GitLab, Bitbucket, Gerrit, Perforce, Azure DevOps) into a unified, deterministic search corpus, available to humans through the UI and to agents through these surfaces: MCP Server. Exposes SCIP-powered code intelligence to any MCP-compatible agent. Instead of approximate, embedding-based retrieval, the agent gets cross-repo, deterministic results: exact symbol definitions, callsites, and implementers where code intelligence is indexed and supported. Stripe's Alistair Gray, quoted on the Sourcegraph homepage: "Minions are connected to MCP. This is how they gather context: internal docs, ticket details, build statuses, and code intelligence via Sourcegraph search." Deep Search. Natural language code search with citations. Engineers, support, and go-to-market teams ask questions like "How are versioned docs deployed?" and get grounded answers. Code Search. The deterministic, exact, exhaustive engine underneath everything. Trusted by enterprise teams for over a decade, the reason searches can return exact call sites rather than a plausible-looking subset. Code Insights. Tracks migration adoption, framework usage, and risk patterns over time. Useful for oversight on agent rollouts. Agentic Migrations (Experimental). A purpose-built workflow for cross-cutting changes: rolling out a security fix, deprecating a function, moving framework versions across every repo at once. The direct antidote to the 80% problem. None of this competes with Claude Code, Codex, Gemini CLI, or any other agent. It sits underneath them. As Sourcegraph's post on why code search at scale is essential lays out, coding agents optimize for the code you're writing, while enterprises need visibility into all the existing code. Teams that want a unified product use Amp; teams standardized on Claude Code, Codex, or Gemini CLI can wire in the MCP server without changing the agent. Honesty caveat: on a small single-repo project, you don't need this codebase-wide context. Sourcegraph's value shows up the moment your repo count crosses 1 and accelerates towards 100. Conclusion: Agentic Coding Done Right Means Engineering Discipline at Scale The narrative that AI is going to make engineering disappear keeps getting louder, and the day-to-day reality keeps drifting in a different direction. Coding agents write code that ships, but they miss in predictable ways, and at Big Code scale, those misses compound into outages and silent debt unless you build context infrastructure underneath them. The takeaway: agentic coding's productivity ceiling is set by context, not model quality. Pick the agent your team trusts, then invest the same energy in the substrate that gives it a complete view of the codebase. That's where the cost savings, the speed gains, and the difference between merging cleanly and breaking three downstream services actually live. If you're past the toy-repo phase, see how Sourcegraph gives coding agents complete codebase context and book a demo to see it wired into your stack. Frequently Asked Questions What is agentic code? Agentic code is code produced by an autonomous AI agent that plans, writes code, and tests software through a tool-using loop, rather than by a developer accepting one suggestion at a time. The term covers both the resulting code and the practice of building software this way. What is the difference between vibe coding and agentic coding? Vibe coding is a casual posture (trust the model, ship if it runs, skip the diff) coined by Andrej Karpathy for prototypes. Agentic coding is an engineering practice: an autonomous agent runs a "plan, write, test, refine" loop, and a human reviews the diff against real acceptance criteria. Vibe coding optimizes for speed on disposable work. Agentic coding optimizes for production code with the same code review discipline you'd apply to a human pull request. What is the 80% problem in agentic coding? The 80% problem is the pattern where coding agents reliably complete the visible 80% of a task and silently miss the 20% outside their context window: cross-cutting changes, sibling repos, auth middleware, audit logging, integration tests, frontend guards. The fix isn't a better model; it's giving the agent a deterministic, repository-wide view of the codebase so the invisible 20% becomes visible before it ships. How expensive is agentic coding? Costs break into model inference, retrieval, and indexing infrastructure, and human review time. Inference is the most volatile line item because agents make many calls per task during the refine loop. Teams report wide variance depending on context quality: with the right context, agents retry less and finish faster. Benchmarks like Sourcegraph's CodeScaleBench focus on how that retrieval quality plays out on large-codebase and multi-repo tasks. Unblock your organization. Ship faster. With Sourcegraph, the code understanding platform for enterprise. Schedule a demo ### On-screen text ### Every link captured - Documentation -> https://sourcegraph.com/docs - February 2025 tweet -> https://x.com/karpathy/status/1886192184808149383 - Collins Dictionary's Word of the Year -> https://blog.collinsdictionary.com/language-lovers/collins-word-of-the-year-2025-ai-meets-authenticity-as-society-shifts/ - agentic engineering -> https://simonwillison.net/guides/agentic-engineering-patterns/what-is-agentic-engineering/ - agentic coding generating more code -> https://sourcegraph.com/blog/agentic-coding-is-creating-more-code-more-code-is-creating-a-bigger-need-for-code-search - brute squad -> https://sourcegraph.com/blog/the-brute-squad - Amp -> https://ampcode.com/ - why code search at scale is essential -> https://sourcegraph.com/blog/why-code-search-at-scale-is-essential-when-you-grow-beyond-one-repository - Sourcegraph gives coding agents complete codebase context -> https://sourcegraph.com/ - book a demo -> https://sourcegraph.com/contact/request-info

### Spoken transcript Simon Willison’s Weblog Subscribe Sponsored by: Depot — AI agents write code in seconds. CI shouldn't make them wait minutes. Try Depot CI Writing about Agentic Engineering Patterns 23rd February 2026 I’ve started a new project to collect and document Agentic Engineering Patterns—coding practices and patterns to help get the best results out of this new era of coding agent development we find ourselves entering. I’m using Agentic Engineering to refer to building software using coding agents—tools like Claude Code and OpenAI Codex, where the defining feature is that they can both generate and execute code—allowing them to test that code and iterate on it independently of turn-by-turn guidance from their human supervisor. I think of vibe coding using its original definition of coding where you pay no attention to the code at all, which today is often associated with non-programmers using LLMs to write code. Agentic Engineering represents the other end of the scale: professional software engineers using coding agents to improve and accelerate their work by amplifying their existing expertise. There is so much to learn and explore about this new discipline! I’ve already published a lot under my ai-assisted-programming tag (345 posts and counting) but that’s been relatively unstructured. My new goal is to produce something that helps answer the question “how do I get good results out of this stuff” all in one place. I’ll be developing and growing this project here on my blog as a series of chapter-shaped patterns, loosely inspired by the format popularized by Design Patterns: Elements of Reusable Object-Oriented Software back in 1994. I published the first two chapters today: Writing code is cheap now talks about the central challenge of agentic engineering: the cost to churn out initial working code has dropped to almost nothing, how does that impact our existing intuitions about how we work, both individually and as a team? Red/green TDD describes how test-first development helps agents write more succinct and reliable code with minimal extra prompting. I hope to add more chapters at a rate of 1-2 a week. I don’t really know when I’ll stop, there’s a lot to cover! Written by me, not by an LLM I have a strong personal policy of not publishing AI-generated writing under my own name. That policy will hold true for Agentic Engineering Patterns as well. I’ll be using LLMs for proofreading and fleshing out example code and all manner of other side-tasks, but the words you read here will be my own. Chapters and Guides Agentic Engineering Patterns isn’t exactly a book, but it’s kind of book-shaped. I’ll be publishing it on my site using a new shape of content I’m calling a guide. A guide is a collection of chapters, where each chapter is effectively a blog post with a less prominent date that’s designed to be updated over time, not frozen at the point of first publication. Guides and chapters are my answer to the challenge of publishing “evergreen” content on a blog. I’ve been trying to find a way to do this for a while now. This feels like a format that might stick. If you’re interested in the implementation you can find the code in the Guide, Chapter and ChapterChange models and the associated Django views, almost all of which was written by Claude Opus 4.6 running in Claude Code for web accessed via my iPhone. Posted 23rd February 2026 at 5:43 pm · Follow me on Mastodon, Bluesky, Twitter or subscribe to my newsletter More recent articles Porting the Moebius 0.2B image inpainting model to run in the browser with Claude Code - 22nd June 2026 sqlite-utils 4.0rc1 adds migrations and nested transactions - 21st June 2026 Datasette Apps: Host custom HTML applications inside Datasette - 18th June 2026 This is Writing about Agentic Engineering Patterns by Simon Willison, posted on 23rd February 2026. blogging 123 design-patterns 18 projects 540 writing 31 ai 2,088 generative-ai 1,845 llms 1,813 ai-assisted-programming 391 vibe-coding 92 coding-agents 212 agentic-engineering 52 site-upgrades 26 Next:I vibe coded my dream macOS presentation app Previous:Adding TILs, releases, museums, tools and research to my blog Monthly briefing Sponsor me for $10/month and get a curated email digest of the month's most important LLM developments. Pay me to send you less! Sponsor & subscribe Disclosures Colophon © 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 --- ### On-screen text ### Every link captured - Try Depot CI -> https://10xn.link/simon-depot - Agentic Engineering Patterns -> https://simonwillison.net/guides/agentic-engineering-patterns/ - original definition -> https://simonwillison.net/2025/Mar/19/vibe-coding/ - under my ai-assisted-programming tag -> https://simonwillison.net/tags/ai-assisted-programming/ - Design Patterns: Elements of Reusable Object-Oriented Software -> https://en.wikipedia.org/wiki/Design_Patterns - Writing code is cheap now -> https://simonwillison.net/guides/agentic-engineering-patterns/code-is-cheap/ - Red/green TDD -> https://simonwillison.net/guides/agentic-engineering-patterns/red-green-tdd/ - Guide -> https://github.com/simonw/simonwillisonblog/blob/b9cd41a0ac4a232b2a6c90ca3fff9ae465263b02/blog/models.py#L262-L280 - Chapter -> https://github.com/simonw/simonwillisonblog/blob/b9cd41a0ac4a232b2a6c90ca3fff9ae465263b02/blog/models.py#L349-L405 - ChapterChange -> https://github.com/simonw/simonwillisonblog/blob/b9cd41a0ac4a232b2a6c90ca3fff9ae465263b02/blog/models.py#L408-L423 - associated Django views -> https://github.com/simonw/simonwillisonblog/blob/b9cd41a0ac4a232b2a6c90ca3fff9ae465263b02/blog/views.py#L775-L923 - Mastodon -> https://fedi.simonwillison.net/@simon - Bluesky -> https://bsky.app/profile/simonwillison.net - Twitter -> https://twitter.com/simonw - subscribe to my newsletter -> https://simonwillison.net/about/#subscribe - Sponsor & subscribe -> https://github.com/sponsors/simonw/ - code-is-cheap chapter -> https://simonwillison.net/guides/agentic-engineering-patterns/code-is-cheap/ - red-green-tdd chapter -> https://simonwillison.net/guides/agentic-engineering-patterns/red-green-tdd/

### Spoken transcript I've built enough agents now to confidently say that broad agents and niche skills is the meta. If you're trying to automate content, you probably don't need thumbnail agent, a description agent, ex-agent, a LinkedIn agent. That sounds really organized on paper, but in practice, it is a building and maintenance nightmare. Now you've got six prompts to maintain, six sets of tool permissions to maintain, and six places for things to go wrong. And they will. The better pattern that I've found is just one broad agent that just handles everything for content. And then this is the important part, down skills, a skill for copywriting, a skill for newsletter writing, a skill for YouTube thumbnails. The models are so good now that over-specializing the agents just makes the whole system more fragile as a whole. You do not need like 40 different tiny agents all handling tiny little processes. You really only need like a few, maybe even one really capable agent and then like really, really well thought out, well put together skills. Happy clotting, ladies and gentlemen. ### On-screen text you dont need 40 agents... you need 1. confidently is the content you probably a thumbnail agent a description a Linkedin agent that sounds really organized on paper in practice a building maintenance nightmare and they will the better pattern found is just one broad agent that just everything for content this is the important a library of really thought Newsletter newsletter Newsletter Youtube Thumbnail a skill for the models so good that over the agents just makes the whole system more fragile a whole you do not need like different tiny agents all handling tiny little you really only need like a few maybe even one really capable and then like really well thought out well put together skills happy Clauding ladies and gentlemen

### Spoken transcript I use around 10 AI tools for 90% of my work, and each one excels in one specific area. But figuring out which tool works best for what task usually takes months of trial and error. So I'll share the one thing each tool does better than alternatives, so you walk away with a clear mental model for when to use what. I've grouped these tools into four categories across a two-part series. There's just too much to cover. This video covers everyday and specialist AI, while part two covers the remaining two categories. Let's get started. Kicking things off with everyday AI. These are your general purpose chatbots, ChatGPT, Gemini, and Claude. And while they seem interchangeable, their quote-unquote moats, the specific things they do best, have actually become quite distinct. Starting with the OG, ChatGPT. While Gemini and Claude are arguably just as capable in raw power, ChatGPT still holds the crown in one area. It's the most obedient model. In plain English, a complex checklist. Other models might be just as smart, but give them a lengthy set of instructions and they'll sometimes skip a step or decide they know better. If you want proof of this, just ask each model to optimize a rough prompt for itself. ChatGPT will generate a noticeably longer and more detailed prompt because it knows it can handle the complexity. And if you run that optimized ChatGPT prompt through both ChatGPT and Gemini, for example, you'll notice two things. First, ChatGPT thinks longer checking every requirement, and it follows each instruction to the letter. Gemini, on the other hand, often takes shortcuts. Pro tip, I shared the exact prompt optimizer in the essential power prompts template linked below, but you can test this yourself with something as simple as, optimize this prompt for ChatGPT, insert model number here, here's my rough prompt. Diving into a real world example, I gave both ChatGPT and Gemini the same complex prompt, a hiring rubric with a dozen requirements. Every single one. Gemini's output looked right at first glance, but when I checked it against my original list, it had quietly dropped a few rules. That's the key difference. ChatGPT doesn't decide which instructions matter. It just follows them. Here's a second, simpler example. Sometimes when you explicitly tell Gemini to search the web, it just doesn't, which is wild since Gemini and Google search are both Google products, right? Whereas with ChatGPT, when you enable web search, single time. I know this is a small example, but it's downstream from ChachiBT's core superpower. Obedience means you can trust the behavior you ask for. So as a rule of thumb, if a task has a lot of moving parts and getting one wrong breaks the whole thing, start with ChachiBT. Next up, Gemini. Where ChachiBT wins on obedience, Gemini wins on multi-modality. In plain English, Gemini is able to process a massive amount of mixed media, video, audio, images, and text natively. Taking a look at this table, we see that only Gemini can handle all four types of media natively. It's able to quote-unquote listen to audio and quote-unquote watch videos, while ChatGPT and Cloud use roundabout ways to access that information. What's more, Gemini's massive 1 million token context window means it can handle large video recordings, hour-long audio recordings, full slide decks, all together that would literally choke other models. If you watch my latest Gemini video, recorded a messy walkthrough of myself completing a task, uploading that video onto Gemini, and asking Gemini to turn it into a ready-to-use SOP with perfect formatting, which is an example of Gemini ingesting video and turning it into text. Now, let's take that a step further. Imagine you just finished a weekly meeting. You have a video recording of the call, a 20-slide deck, and a photo of a messy whiteboard session. You can upload all three and ask Gemini to summarize what was discussed, follow-up email. Gemini is the only tool that can synthesize all three in one go. All that said, I have to point out that Gemini's raw reasoning capabilities sometimes feels slightly behind ChatGPT. But when the task involves video, audio, or massive files, the trade-off is obviously worth it. Speaking of matching the right tool to the task, today's sponsor HubSpot put together a free guide called the AI Productivity Stack that covers 50 tools organized by use case. are my personal favorites, your workflow probably needs something different. Maybe you're in marketing and need SEO specific tools, or you manage a team and want to build automated workflows with reliable AI. This guide breaks down tools across business functions like research, design, and marketing. And for each tool, it shows you the best use case, key features, pricing, and a step-by-step workflow. What I found most useful is the decision logic at the end of each section. use perplexity versus Claude versus Humada based on what you're actually trying to do. It's a great way to quickly understand what each tool does well. I'll leave a link to this free guide down below. Thank you, HubSpot, for sponsoring this video. Rounding out the everyday AI category, Claude. Claude's superpower is producing higher quality first drafts than the other models. In plain English, that means Claude's first attempt is usually closer to done. This superpower shows up in two areas. First, coding. Here's a fun fact. of Gemini beat the older version of Claude in every single benchmark score, except for the coding one, which is crazy. So obviously Anthropic has figured out something related to coding the others haven't. And in practice, developers universally agree that Claude writes functional code on the first try more consistently than alternatives. Here's a real world example. I needed to bulk export conversations from a customer service platform, developers could do it. I described the problem, and Claude not only gave me step-by-step instructions, but also wrote a script in Go that worked on the first try. I don't even know what Go is, nor can I write code. Another example, I asked all three models to turn a static image into an interactive chart, and Claude performed the best on the first try. So basically, anything that requires generating working code tends to favor Claude. Pro tip, when it comes to diagrams, you can ask Claude to generate mermaid code, into tools like Excalibur to get clean visuals in minutes. Area two, polishing copy. Beyond code, Claude produces written drafts that sound human and need fewer revisions. When you need to tighten an argument or match a specific voice, Claude just gets it. Put simply, it's exceptionally good at style matching. Once you share examples of your existing work, it replicates your tone almost perfectly. When I was in corporate, I'd share previous documents so Claude could replicate that voice across presentations And now as a creator, I feed it my existing YouTube scripts to help refine new drafts. At this point, you might be wondering how I use all three everyday AI tools together. In a nutshell, ChatGPT or Gemini usually handles the beginning of my work. Ideation, research, drafting the outline of a presentation. Cloud that handles the last mile, turning that rough output into something I'm ready to present or publish. Quick note on Grok. A lot of people ask why I don't use it. It's actually very simple. X firehose, right? So it's the best option for people who need to analyze breaking news events in real time. I never needed that. And as a rule of thumb, we should never use tools just for the sake of using tools. We should only add them to our toolkit when they solve an actual problem we have. Here's a quick recap of the three models and when to use them. And if you're wondering whether you need all three, the short answer is no. Most people should stick with a paid version of ChatGPT and get really good at it. But if you can afford multiple subscriptions and your workflow can take advantage of their individual powers mix and match as needed. Fun fact, according to this study on open router data, models from different labs like ChatGPT and Gemini expand the pie of AI use cases precisely because they excel at different things. Onto the second category, specialist AI. Before diving in, let's clear up a very common misconception. Tools like Perplexity are not foundational models. Here's a simple visual. OpenAI, a frontier AI lab, develops the GPT family of models. the user-friendly app layer. Perplexity is different. It fine-tunes existing foundational models for speed and accuracy and is optimized for search. Their own Sonar model, for example, is just a fine-tuned version of Meta's open-weight Lama model. So on that note, Perplexity's superpower is finding accurate information fast. In plain English, the general purpose chatbots are built for reasoning. You use them to help you think, brainstorm, or write a draft. for fetching. You need a specific fact and you need it now. Starting off with a simple real life example. I used chachibee tea to plan a trip to Japan with my brother because that is a creative task. It requires weighing trade-offs, building a narrative, and for that kind of task, I'm happy to wait while the model thinks. But when I need grab and go information, like whether a specific restaurant is foreigner friendly because we don't speak Japanese, I'd want perplexity to give me accurate and update information within seconds. Second example, going back to how I use AI tools, let's say Gemini or ChatGPT helps me brainstorm and structure my newsletter. Claude produces the final draft. Perplexity in this case is the search scalpel that verifies information like whether Gemini's context window is 1 million or 2 million tokens. In case you're curious, consumers get 1 million, enterprises get 2 million. Pro tip, you can use Google style search operators like site colon reddit.com to narrow your results to a specific source. I have an entire video on the most useful Google search So I'll link that down below. As a rule of thumb, think of Perplexity as a replacement for Google AI mode. They're both for fetching information and not as a replacement for general purpose chatbots. Actually, let me know if you want an entire video breaking down the AI search apps like Perplexity, Google Search, Google AI Overviews, Google AI Mode, because they're all made for different things. Rounding out specialist AI, Notebook LM's superpower is that it only answers from the sources you give it, meaning it won't make things up. Think of it like a walled garden. You upload your sources and Notebook LM answers questions using only those documents. It can't really hallucinate because it has no outside knowledge to draw from. Going back to the visual around how perplexity is optimized for search, Notebook LM uses a fine-tuned Google Gemini model that minimizes hallucinations. For instance, when I was at Google before publishing marketing materials, I would upload the final draft alongside the source documents and ask Notebook LM if the draft made any claims that contradicted the sources. catch these tiny discrepancies other AI might have missed. I use a similar workflow today for my videos. Before I start filming, I upload my script and all my research into Notebook LM and ask it to flag anything not directly supported by the source material. The obvious caveat here is that the output is only as good as the sources we give it. So if the sources are incorrect, Notebook LM is going to be confidently incorrect. So as a rule of thumb, if the accuracy matters more than creativity and you have source materials check against, use Notebook LN. There are a few more specialist AI tools I use, but didn't make this list because I don't use them every day, but to quickly go through them, Gamma for presentations, 11 Labs for voice cloning, Zapier and NAN for automation, and Excalibur and Napkin AI for quick visuals. As a reminder, I'll cover the remaining two categories in part two, so keep an eye out for that. See you on the next video. In the meantime, have a great one. ### On-screen text Category 4 Category 1 Category 3 Category 2 Creative Al Specialist Al Productivity Al Everyday Al EVERYDAY AI why I use Gemini and Claude are arguably on par when it comes to raw capability, but ChatGPT is the most obedient model. Forgot about this. Forgot about this streeger first 2 Lines (the "hock" befere "-me asce?) featured sectioed Your sole function is to act as a prompt opomizer." 5i Quick wins Checklist Gemini blended theing Share Why it matters: The current first lines read like a career anecdote, not a promise to the vistor in the first 2 seconds Why it matters: Featured is your conversion block, yours currently shows peeric "Link" ties Why it matters: The role statement is values-based but not scarnable of proof-driven. Skills: align the top 3 with what you want to be hired/booked for tended thinking Gemini To refere these edits furthe, I would ideaty need to know Primary Goat: is the prievity cumenty newsletter signups, course sales for A System sademy," or corporate workshop inquirie Missing "Abeur Content: What are the speckie pedf metics or team aides yo artioned in the truscaned "see more" section? Newsletter Stars: Do you have a specific Ourrect Reader Count" or "Time Saved merric for the Sushie Labs newsletter? Executive Summary Park Change The Attor Mosk Mes What's on the agenda today? Optimize this prompt for ChatGPT [insert model number]. Output in a code block. Here's my original prompt: Explain Google's entire Al tech stack in plain English. Extended thinking v we co you wrayss of weaknesses of areas with outicer esigned to be used in a follow-up interview or reference check t Cand date Evaluation Repor Criterion Cceeds Expectations - Very strong, authente slgiment with you his a wive east want i teach o seracy/productivity and adt lon for your structured, system Edended thinking Gemini refuses to use search tool Good to see you, Jeff. Extended thinking Search the superpower Obedience means you can trust the behavior you asked for. the superpower Obedience means you can trust the behavior you asked for. rule of thumb the task has a lot of moving parts and getting one wrong breaks the whole thing, start with ChatGPT. EVERYDAY AI Native Multimodal Capabilities Claude Gemini processes text tokens nate. Text natively built to see and reason Image processes visual inputs natively. about pixels. Audio Partially - handles real-time speech No - not native; converts audio to text natively listens to audio streams natively, but relies on transcription for before processing. and tres without converang to text. No - not native; relies on breaking Video No - not native; simulates watching by natively watches video with sampling individual frames. video into static image frames. temporal reasoning, just like a human. Native Multimodal Capabilities Claude Gemini processes text tokens natively Text natively built to see and reason Image processes visual inputs natively. about pixels. Audio Partially - handles real-time speech natively listens to audio streams No - not nauve; converts audio to text before processing. and files without converting to text. natively, out reles on transcription Tor No - not native; simulates watching by Video No - not native; relies on breaking natively rideo with sampling individual frames. video into static image frames. temporal reasoning, just like a human. Native Multimodal Capabilities Gemini Claude Text processes text tokens natively processes visual inputs natively. natively built to see and reason Image about pixels. Audio Partially - handles real-time speech verts audio to text natively listens to audio streams No - not native; natively, but relies on transcription for before processing. and files without converting to text. with No - not native; natively Video No tive; relies on breaking sampling individual frames. video into static image frames. temporal reasoning, just like a human. Jeff Where should we start? Product..vity Stack Your task is to watch this video, analyze and deconstruct its contents, both text and visual, and to understand what it's about. | Write anything Create image Create video Explore visually Help me learn Boost my day Master Gemini 3.0 for Work in 12 Minutes (2026) Jeff Su Link Down Below MULTIMODALITY IN PRACTICE Jeff Where should we start? Product...vity Stack Analyze all attached and distill key learnings. Create image Write anything Create video Explore visually Boost my day Help me learn Gemini's raw reasoning capabilities sometimes feels slightly behind ChatGPT. when the task involves video, audio, or massive files, the tradeoff is obviously The Al Productivity Stack: Tools for Output motion In partnership with Media Description Category Tool Al-powered social media marketing Marketing & and analytics Social Media I-driven social media monitoring and brand management. Al-powered social media ad targeting Meta Al and automation. No-code automation and Al Business Automation task automation. powered creative and business A4Ooe Sense workfiow optimization. Task Management & Organization Managing tasks effectively is essential for staying productive. manual scheduling and prioritization can quickly become overwhelming. The right tools can help you automate workflows. streamine your schedule, and keep proiects on track with minimal Marketing and Social Media Buffer Marketing and Social Media Creating and managing content at scale requires efficiency and strategy. These tools streamline content creation, scheduling, and performance analysis, helping businesses reach the right audience with less effort. Whether automating social media posts, optimizing for search, or generating high-performing ad creatives, they make boosting engagement Buffer Best Use Case: Al-Powered Social Media Post Pricing: Free basic plan; paid plans start at Creation & Scheduling How to Use It in a Productivity Workflow. Draft a social media post and use Al suggestions for copy improvements Twitter, LinkedIn, and Pinterest. Hashtag recommendations to boost reach Analyze post performance and adjust content strategy based on Al insights. Download below for free why I use first drafts than the other models. simply, its first attempt is usually closer to "done." Coude Sonnet Description seanh are mite verttd No book| Reasoning vending Bench 2 Active ~ Conversation History Export Request @ Chat Overview days agal Help Scout Team, Good day! We would like to request a ful export of the conversation history associated with our inbar days ago) Cartery, you can only export your reporting datin-app. Otherwise, you'd need lo leverage our bak all reteve mol other data, such in full conversations, or our dachAl for your Does articles. See more here: Export account data. Using the API does require some developer time and skill, but it will do just what you need. Here's what wore able to export for you om the back and of Hel Scout: Saved Repses Customer contact information by mailbox or sag, or a full list Pease it ese know 11 can send any of this over to yout Find your Mallbox LD by looking at the Ust when you're Viewing your laboo. It's the The script will asthenticate, fetch all conversatioes pull the fut theead content tor cach cne. What to expect: The script processes abost 25 comenacions per second because of Help pectress updates the whole time something goes wrong: Copy the exact erece menage and paste it here. Most lowes and typos in the credentials or Go not being installed correct: Top 6 Al coding assistants by size of organization swapping in the coerect mambers i a cae-ine edit in the data array at the topAmber of unem You could adjust the animation duration (cursently 1500ms), change the colce palette in than all together. Let me know If any of those treeaks would be useful coding chart Gemini Preview Data Visualization Turn this static image into an animated chart. turned that static chart image into a fuly Replay Animation (Audrey te teara thembef-bandover te Lyndst1 =, Refine my draft based on my previous newsletters. the artifacts feature Just cutput as normal Habits That Will Take You From ALterate to A-Native Most professionais use Al as a "iterace" level, knowing welch tools to use and started working at Google and was fooed to use only Google Workspace tools one missing cature drove me absolutely cray.. That's concrete personal, and immedaly relatic starting with a specific moment where you reallzed the difference between literate and Let's pet started needs to po this doesn't match your established tone Four newsletters are direct and get into the content without cheerleader energy: Corporate language creeping in Imegrate artificial Intellizence directle Inso your dally operatkins" sounds like a press relase Same with opsintee your planning. Your stytels Emoll usage Your Google Calendar plece uses exactly one emoll at the very end (2) asa sign-of. The current draft has * callour bones throughout. You could Claude produces written drafts that sound human and need fewer revisions. When you need to tighten an argument or match a specific voice, Claude just gets it. Once you share examples of your existing work, it replicates Now as a creator, I feed it my existing YouTube scripts to help refine new drafts. BUSINESS PRESENTATION IDEATION, RESEARCH, OUTLINE Search KR Why are people so mean on twi why are people so mean asine why are people so mean on redalt why are pecole so mean on roblas a why are peccle so mean on social media why are people so mean on the Internet Grok's superpower is its direct access to the Twitter/X firehose, so it's the best option for people who need to analyze breaking news events in real time. Searching the web 18 results Grok's superpower is its direct access to the Twitter/X firehose, so it's the best option for people who need to analyze breaking news events in real time. We should never use tools just for the sake of using tools, we only add them to our toolkit EVERYDAY Al the task has a lot of moving parts and getting wrong breaks the whole thing, start with ChatGPT. Gemini: If the input is video, audio, or messy files, Gemini is the only everyday Al that processes it natively. Claude: If you need working code or polished copy on the first try, start with Claude. aude Sonnet 3.7 AMni Flash 2.0 Tokens (Millions' Gemini 2.5 Models from different labs expand the pie of Al use cases instead of just "stealing" market share from each other. FOUNDATIONAL APPLICATION FOUNDATIONAL Sonar Gemini x Perplexity Al Best Sonar anything. Type @ for mentions and PERPLEXITY SPECIALIST AI why I use PERPLEXITY Perplexity's superpower is finding accurate information, fast. The general purpose chatbots are built for reasoning, while Perplexity is built for fetching information. wink you even conte and ecle color loudy rome kink thorne the toothim maturants you Morn NEWSLETTER WORKFLOW Final Draft Brainstorm + Outline "Search scalpel" public release, enterprise-specific configurations, and the capabilities of the previous Gemini 1.5 model. The Breakdown: What is the "Truth"? Enterprise / Preview Options Model Version Standard Context Window Million Tokens Gemini 3.0 Pro Up to 2 Million Tokens for specific enterprise configurations Gemini 3.0 Million Tokens gemini +1 Flash Million Tokens Gemini 1.5 Pro widely available for developers a follow-up perplexity pro Why are we doing nothing about the Epstein list? People also ask ! Google Tips that will Save You Hundreds of Hours! Jeff Su Link Down Below PERPLEXITY the superpower Finding accurate information, fast. rule of thumb Think of Perplexity as a replacement for Google Al Mode, not as a replacement for Everyday Al. Overview SPECIALIST AI why I use superpower is that it only answers from the sources you give it, meaning it won't make things up. World Wide Web APPLICATION FOUNDATIONAL Gemini (fine-tuned to minimize hallucinations) for an in-depth report and now sources! PMax Notebook Search the web for new sources The provided sources describe Performance Max, an automated artificial intelligence to distribute advertisements across all Goog Fast Research v videos, and text, the system autonomously manages bidding an objectives such as lead generation or sales. While this approac Select all sources adjustments and expands brand reach, it requires high-quality granular transparency. Experts suggest that while the Al-driven Performance Max Comms Doc.pdf must maintain clear conversion tracking and realistic budget goo documentation positions this tool as a powerful, all-in-one solut Rough ideas.pdf presence with less day-to-day oversight. Save to note How does Performance Max utilize artificial intelligence to auto Chat Fact-check my sough idea based on the COmme Doe Pavieving the Key Clemer for besear of the pl factually eunurate and wigh wet with the core det ancy of the bi Your Draft: You l Confirmed. The Comms Doc describes it as One mega-campaian that can appear almost anywhere Google can show ads," specifically listing Search, Shopping, YouTube, Display, Discover, Gmail, and Maps 2, 3. Shan Chat Video 195 | Gemini fies AP Gemini APt Google Al for Devel. Denie 3 for developers New meetig h / Cere 3.0. Voldes hoe Werk Gemini thinking - Ganini At- Googh Nilo..V Prompt design sraleges GemiiA/t Go. V fased on the provided sauces, here is a fact-check of your sorer Miser Germin 3.0 Nr Work in 2 Minute Search the web for new sources the documentation. However there are specific dura points-particularly benchmark names and percentages -thal Timeline C2034) Accurse context, Gemini 3 was released in November 3025 1, 2.-A video sded "Maste Genin. (ar is cheonciogically appropriate Preview s Claim Gerin 3 processes afe, viled, and lages Aspecher nather than breaking them down a screenshot Gemini thing - GeniAl GoNNa D and "vides masoning". it mplicity not that Gemini can handle input data including ted, images, and sudo "at Case (Video Analysis: Supported. The sources confe the model can "anayae vece of your pickleball Media resolution Germie All Goople AlL V nich kely went where star mary wernery stone care Prompt design strategies GeniAl Go. V The output is The output is only as good as the sources we upload. So if the sources are incorrect, the the superpower Lowest hallucination rate of any Al tool. rule of thumb accuracy matters more than creativity, and you have source materials to check against, use NotebookLM. Category 3 Category 4 Creative Al Productivity Al ### Video description & links ⚡️ HubSpot’s Free Guide: https://clickhubspot.com/5c4cca Every AI tool excels at one thing. #ChatGPT follows complex instructions without dropping steps. #Gemini handles video, audio, and massive files other models can't process. #Claude produces working code and polished copy on the first try. Perplexity fetches accurate information in seconds. NotebookLM answers only from your sources, so it can't hallucinate. This video breaks down when to use each tool based on three years of daily use, first at Google and now as a creator. Part 1 covers Everyday AI and Specialist AI. Part 2 covers Creative and Productivity AI. *TIMESTAMPS* 00:00 The AI Tools that Drive 90% of My Results 00:32 Everyday AI 00:47 ChatGPT 02:46 Google Gemini 05:15 Claude 07:45 Recap: Everyday AI 08:14 Specialist AI 08:47 Perplexity 10:20 NotebookLM *RESOURCES MENTIONED* My Essential Power Prompts template: https://www.notion.com/templates/essential-power-prompts-jeff-su Google search operators video: https://youtu.be/DIuo_QL4sAQ Blogpost: https://www.jeffsu.org/10-ai-tools-drive-most-results/ *BUILD A POWERFUL WORKFLOW* 🦾 AI Systems Academy - https://systemsacademy.ai/?utm_source=youtube&utm_medium=video&utm_campaign=197 📈 The Workspace Academy - https://academy.jeffsu.org/workspace-academy?utm_source=youtube&utm_medium=video&utm_campaign=197 ✍️ My Notion Command Center - https://www.pressplay.cc/link/s/DE1C4C50 *BE MY FRIEND:* 📧 Subscribe to my newsletter - https://www.jeffsu.org/newsletter/?utm_source=youtube&utm_medium=video&utm_campaign=description 📸 Instagram - https://instagram.com/j.sushie 🤝 LinkedIn - https://www.linkedin.com/in/jsu05/ *MY FAVORITE GEAR* 🎬 My YouTube Gear - https://www.jeffsu.org/yt-gear/ 🎒 Everyday Carry - https://www.jeffsu.org/my-edc/ #ai

### Spoken transcript Every creator that was making N8N content six months ago is now making Claude Code content. Is N8N dead? I actually use both. I moved a lot of things into Claude Code, but I kept some crucial things in N8N. And I think it's so important that you know when to use which for what. Otherwise, you're going to end up building your entire business infrastructure completely wrong. See these types of N8N agent swarm builds. These are dead. Finished. Never, ever build another one of these ever again. These complicated builds with agents talking to other agents and tools connected, that is literally Claude Code. Open up Claude Code. customize and connect up any app that your heart desires. Now you have a 10 times more powerful version than this out of the box. And then simple day-to-day tasks that you use to automate with NNN, things like transcribing YouTube videos, that is completely dead too. These type of workflows, they need to become Claude's skills. You can see here, I've created a transcribe skill and I can give it any YouTube URL, click enter, and it literally will just transcribe that for me. But there's still a certain type of workflow that should never enter Claude code. background processes for your business that run the same way every time and have a set trigger and a set outcome. This one, for example, is for customer support. They get a new email in, they have to do all this process to find data, summarize things, and then send them back a set type of email. That should run in the background, never break, and none of the employees and the founder should ever even see it happening. If it runs in the background, it's rigid, step-by-step, same things happen each time, it's probably better off as an N8N workflow. If it's an agent to help you with different roles in your business, like marketing or sales, team or an open call. If it's a manual day-to-day process that's part of your work, like transcribing YouTube videos, it's better off as a skill. ### On-screen text should I automate things in n8n or claude code? claude code content is n8n dead? actually use both moved a lot of things into claude code kept some crucial things in n8n and I think it's so important that you know when to use which otherwise you're gonna end up building your entire business infrastructure completely wrong Inactive Agent Swarm Editor Personal Waiting for trigger event see these types of . wetter erft reet script the Brand Writer ter "kt with Romy." Vars en 3 hms opplans - pick the ne that foeke most Iatro 1 - Call Out the Orestes Ploot The Bul Hypo Angle Inactive Agent Swarm Editor Personal Waiting for trigger event agent swarm builds star these are dead Waiting for trigger even these are dead has 5- Many tin Bitu Agent Swarm Personal Inactive Editor finished never ever build another one of these ever again nate herk nan agent Agent Swarm Waiting for trigger event Bullding an Al Agent Swarm in nan Just Got So Easy lite Hark LAl Automatm s Share Save Thanks these complicated builds Bulding an Al Agent Swarm Free M Resources C Cents D Web L Dev D Affilates E Crypto [ Courses nate herk n8n agent Saved Personal Agent Swarm Editor Waiting for trigger event In in nin Just Got So Easy Share save Thanks these complicated builds Bulding an Al Agent Swarm (3.Aces Aces tool Cl Rescorces Do Cients D Web D Dev nate herk n8n agent Saved Inactive Agent Swarm Personal Editor Waiting for trigger event Share Save New sessie chat comon with agents Buldog an Al Agent Saamm x « Woritiows - nên Free M nate herk nen agent Anent Swarl Walting for trigger event Share sare Thanks talking to other agents Automatic × + Free Markdown Ed nate herk nin agent All Star Walting for trigger event Built the Ultimate in nên Just Got So Share Save Thanks and tools connected nate bak sin agail Saved o In Just Out So Easy and tools connected nate here nan agent Agent Swarm a faring Boilt the Waiting for shager event Agent twarm in nin Juet Got So Easy that is literally Claude Code that is literally Claude Code command option Agent 43-mild Waiting for trigger event in n8n Just Got So Easy Thanks /Save Share that is literally Claude Code lock control option command command option Sand Email Mart Unread Waiting Building an Al Agent Swarm in n8n Just Got So Easy Nate Herk | Al Automation e New session Select a folder All projects Connect Claude to your a Anthropic for safety You c Search Personal pligios Frontend design Create dally weather emall workflow Ralph loop General coding session: Draft recitis, summar; Notion Quick disk space usnge scan Notion h Create timestampe and dencription / Stripe Colmact your Notioo w sacch update and no, toes to sand Limose Atlassian «te to customize Intercom Agent Swarm in nin Just Got So Easy Merk All Automation o to customize waning tor wigger rem Chara Sanct a folder Connect Claude to your apps, files, and services. Connectors are built by third parties and reviewed by Anthropic for safety. You can also add a custom connector. Search Sort v Categories v Type v Manage connectors Frontend design Excalidraw interactive for creating interactive hand Ralph loop Manage databases, authentication drawn diagrams in Excaliticaw and storage Notion Hugging Face Project management & collaboration Access the Hugging Face Hub and Stripe Context7 Stripe Up-to date docs for LLMs.and Al Payment processing and financial giving Claude code editors Indeed Stants for jobs on indeed masting transonate Microsoft Learn Clay intran Global and connect up any app Walting for trigger event Share Save Saect a folder Connect Claude to your apps, files, and services. Connectors are built by third parties and reviewed by Anthropic for safety. You can also add a custom connector. Sort v Categories v Type v Manage connectors Search Frontend design Ralph loop Notion Stripe Open Targets Dictatie discovery anal Search Mighta in Clatice giving Claude. Harvey Catalog and Moonir Al Viewer that your heart desires Built the Waiting for trigger event Ultimate view. Swarm in n&n Just Got So Easy Automation o Share Thanks Save now you have command command a 10x more powerful tree Markdown to! nate herk nan agent Saved Inactive • Execution Personal Agent Swarm Walting for triager even Bullding an Al Agent Swarm in nên Just Got So Easy Nate Heck IAl Aeromation Share Save version than this Waiting for trigger event in min Just Got So Lasy Save Share DIll out of the box lock command Share sane Prom your search stop eating fille in Rate T that is completely dead too that is completely dead too these type of workflows they need to become claude skills see here l've created command Share Save 9 clia & Download Code Chat transcribe https://youtu.be/RFI5HIZLx8U?si=VGfJSZ9tHhJLnois Opus 4.6 Local Auto accept edits v click enter option command Save Share Chat and it literally will just there's still a certain type of workflow that should never enter claude code Free Markdown Ed. and it's these really rigid Nutripath Results Automat x Building an Al Agent Swarm i a Dev C Affiliates C Crypto C Courses licrobiome Clinic New Email Received Microsoft Outlook Trigger and have a set trigger Finder Edit File View Window Help Open High this one for example Bolting oo Al Agent saarm is for customer support Bulang an Al Agent Swarm: * CApes & 1605s Do Resources C Clients a Web Cl Dev @ Affiliates Co Crypto M Courses 1a Microbiome Clinic Nutripath Results Automation + Add tag Editor Execute worktlow is for customer support Scrat Building an Al Agent Swarm Editor Microbiome Clinic Execute worktien is for customer support Scrape ArT Youtube Transc Nutripath Results Automat X Fren Markdown Ed. Execute worktio they get a new email in Execute workhow they have to do Resources m Cients Web Execute worktiow all this process Albaba Manufactu.. me Free Markdown to to find data that should run in the background never break and none of the employees and the founder should ever even see it happening it runs in the background rigid step by step same things happen each time probably better off agent to help you with in your business like marketing or sales just set up a claude code agent team open claw a manual day to day process that's part of your work like transcribing better off as a s kill