Stop Prompting AI. Start Managing It. (My Full Setup)
The speaker reveals a structured AI team setup using a local folder system with 30+ specialized agents managed by an orchestrator named Larry. The system is built on real-world business productivity principles, using human team management analogies to make AI delegation more effective. No third-party tools or code are required, and the architecture is designed to be model-agnostic.
Summary
The video presents a comprehensive walkthrough of the speaker's AI team setup, centered around a local folder-based architecture that requires no third-party tools or coding knowledge. The system is built around three foundational agents: Larry (the orchestrator), Nolan (the HR/agent resources manager), and Pax (the senior researcher). Larry's sole job is to translate the owner's requests to the right specialist — a role the speaker compares to his former job as a business analyst bridging business stakeholders and IT teams. Larry is prohibited from doing any actual work himself; he only delegates.
Nolan functions as a human resources equivalent, responsible for 'hiring' new specialist agents when a gap is identified. When Larry determines no existing team member can handle a task, he contacts Nolan, who in turn consults Pax to research what the ideal specialist should look like. Pax then returns a detailed profile, and Nolan creates (hires) the new agent. This workflow mirrors real corporate HR and onboarding processes intentionally, as the speaker argues that AI is trained on human data and therefore responds more effectively when addressed using familiar business productivity language.
The folder structure is explained in detail. The root folder is 'Larry's AI Team,' with Larry's persona stored in a CLAUDE.md file at the root level so Claude can identify him as the orchestrator upon session start. Other team members reside in a 'Team' subfolder with their own CLAUDE.md files. A 'Team Index' file allows Larry to quickly survey available specialists. There is also an 'Owner's Inbox' for work review, an 'Archive' for completed work, a 'BKM' (Business Knowledge Management) folder for long-term team knowledge, and a 'PKM' (Personal Knowledge Management) folder for the owner's personal insights like journaling.
Standard Operating Procedures (SOPs) are stored in the BKM folder and referenced within each agent's CLAUDE.md file rather than embedded directly, keeping individual agent files small and token-efficient. An SOP Index allows Larry to quickly identify available procedures and gaps. A 'Team Inbox' folder allows the owner to hand off files — including scanned documents via a Dropbox integration with a physical scanner — for the team to process.
Session logging is handled via a shared SQLite database stored in the BKM folder, keeping individual agent folders lean while enabling persistent memory across sessions. A hidden .clawd folder houses custom slash commands, such as a 'close chat' command that prompts Larry to log session insights and clean up the inbox before ending a session. A separate SQLite database for the owner's PKM keeps personal and team knowledge cleanly separated. The speaker emphasizes that the entire structure is model-agnostic — it can be migrated to Gemini, Codex, or other models by updating only small configuration files while leaving the core documentation intact. The video closes with a mention of a paid course offering step-by-step guides, prompt sheets, and downloadable folder structures.
Key Insights
- The speaker argues that using human business productivity language with AI agents — terms like 'hiring,' 'HR,' and 'standard operating procedures' — makes them more effective because AI is trained on human data and can better connect to productivity concepts expressed in natural business terminology.
- The speaker designed Larry the orchestrator to be explicitly prohibited from doing any actual work, limiting his role solely to delegation, team organization, and translating the owner's intent to specialists — a design directly modeled on his former role as a corporate business analyst.
- The speaker keeps agent CLAUDE.md files intentionally small and offloads detailed instructions into referenced SOP documents, because smaller files consume fewer tokens and the SOP structure allows switching AI models by updating only minimal configuration while leaving core documentation untouched.
- The speaker separates team knowledge (BKM) and personal knowledge (PKM) into distinct SQLite databases so that team operational knowledge and the owner's personal insights never get mixed, and either component can be independently moved or migrated.
- The speaker stores the entire AI team system inside a local Dropbox folder, which enables device portability, full data ownership, no additional software dependencies, and a direct integration with a physical document scanner that automatically routes scanned invoices and contracts into the team inbox.
Topics
Transcript
[0:00] I know many of you have been waiting for this video. Now is time where I reveal the step-by-step setup of my AI team. And there you will see why it is so efficient and in fact running our whole business with 30 agents in parallel with just one orchestrator in there and all without any code. After this video you perfectly understand the folder structure that I'm using all running on a local folder. No third-party tools needed to get this going. So let's dive right into this. So here we are. This is [0:31] what we will build and this is the foundation of every efficient AI team based on the principles that we build it. So…
Full transcript available for MurmurCast members
Sign Up to AccessMore from ICOR with Tom | AI Productivity
Obsidian is the tool I teach in. Keep the one you have.
The creator addresses criticism about switching to Obsidian after previously promoting other tools, clarifying that his tool-agnostic ICOR methodology remains unchanged and that Obsidian is simply an interface layer for his folder-based system. He emphasizes that the underlying folder structure, AI automation, and productivity principles are identical regardless of which tool visualizes the content.
Claude sucks. And here is why.
This video explores why Claude and other AI models produce inconsistent results, examining how model selection (Haiku vs Sonnet vs Opus vs Claude 3.5 Fable), effort levels, prompt clarity, and system context dramatically affect output quality. The speaker demonstrates these differences through a Venn diagram creation task and advocates for using organized folder systems with documented SOPs and code-based solutions to ensure consistent AI performance.
Claude can now prompt its own SESSIONS. I built the boss.
Claude now has the ability to interact with multiple sessions, allowing a streamlined orchestration of tasks through an agent called Larry. This integration can simplify task management and enhance productivity by enabling a single main session to oversee various sub-sessions or tasks.
I switched off Claude's memory. My folder does it better.
Tom explains why he disabled Claude's native memory feature and instead built a local folder-based system (myPKM) to maintain persistent context with AI agents. He shares an audit revealing systemic issues like contradictory documentation, forgotten rules, and parallel session conflicts, then outlines how to systematically fix these problems using the ICOR productivity methodology.
I do not download invoices anymore. My folder and Claude do it.
The speaker demonstrates how he uses AI (Claude) integrated with his local folder system to automate invoice collection for his accountant, downloading 36 invoices in 5 minutes with minimal manual intervention. He explains his AI team methodology, work streams, and how persistent local knowledge bases enable AI to handle recurring tasks independently without constant re-explanation.