How We Do AI - Two Professionals. One Mindset.
Tom and Paco address community questions about their IICOR productivity learning platform, including an upcoming major update (myICOR 5.0) with 73 new video lessons. They discuss Claude AI memory management challenges, the distinction between personal and business productivity systems, and why foundational productivity frameworks must precede AI implementation.
Summary
The session opens with Tom and Paco explaining they are departing from their usual format to directly address questions and comments from their inner circle community members. They begin by responding to a concern from a member named Julius about feeling overwhelmed by the apparent scale of 290 lessons across 70 courses in the IICOR platform. Tom clarifies that the actual IICOR journey consists of only 73 sequential lessons, and that the larger number reflects supplementary content. He then announces a major upcoming platform update, myICOR 5.0, which includes 73 new video lessons to complement the existing text-based growth assignments, improved cross-referencing between learning resources, and a cleaner integration of YouTube videos, articles, and podcast content directly beneath relevant course lessons.
The hosts explain the core philosophy of IICOR: it is a tool-agnostic productivity framework that covers digital note-taking, personal knowledge management, task management, project management, automation, and now AI. The IICOR framework is visualized as concentric circles representing personal and business productivity areas, and members are encouraged to map their existing tools onto this framework before beginning any courses. Tom emphasizes that most members end up discovering redundancies and gaps in their tool stacks simply by going through this process.
A significant portion of the discussion centers on Claude AI's memory system and how it causes confusion for users. Paco argues that Claude's built-in chat memory is an artifact of an earlier era and recommends users who have committed to Claude Code or Co-work should stop using Claude Chat entirely, as the two systems create conflicting and uncontrollable memory states. Tom demonstrates his 'close chat' command stored in his claude.md file, which triggers his AI agent team to summarize and log the session's conclusions into a structured database — ensuring only the signal, not the noise, is retained. Both hosts advocate for user-controlled local markdown file storage over LLM-managed memory.
The conversation then addresses a question about whether IICOR is designed for individuals or businesses. Tom and Paco firmly state it is built for working professionals, which naturally extends to business contexts. They caution strongly against deploying shared AI systems across teams before individuals have mastered their own personal productivity systems. Paco, who oversees four companies with over 70 staff, describes his approach of focusing AI implementation at the individual level first, leaving existing team project management tools (like ClickUp) unchanged, and allowing the quality and quantity of individual outputs to improve organically. Tom shares his own corporate experience of using personal productivity mastery to gain full visibility over his workload and communicate clearly with management about priorities.
The hosts also tackle a follow-up question about version control and maintaining a single source of truth for shared business AI context. Tom explains that read-only file access — whether via SharePoint, Google Docs, or MCP connections — allows team members to pull consistent context into their personal AI systems without risking data contamination. He notes that their own community uses a similar model, giving inner circle members read-only MCP access to IICOR resources to enrich their personal PKA systems.
The session closes with discussions on Claude's terms of service (warning against using tools like OpenClaw or shared accounts), a reminder that learning transcripts cannot simply be fed to AI as prompts without personal context, and a showcase of the inner circle's event recording library, which includes comprehensive AI-generated summaries, timestamps, prompt templates, and troubleshooting guides. An upcoming workshop on Wednesday is highlighted, with 130 members already signed up.
Key Insights
- Paco argues that Claude's built-in chat memory belongs to the past, claiming that once a user commits to storing context in local markdown files, it makes no sense to return to chat-based memory because the user loses control over what the LLM memorizes and cannot rely on it being relevant or accurate.
- Tom demonstrates a 'close chat' command in his claude.md file that instructs his AI agent team to summarize and log only the conclusions of a session into a structured database, arguing that meetings produce confusion in the middle and only the conclusion matters — so AI logging should capture signal, not noise.
- Paco contends that 95% of businesses fail at AI implementation because they attempt to deploy AI across teams before individuals have solid personal productivity foundations, and that the correct approach is to empower each person's individual work first and let team-level improvement compound from there.
- Tom warns that sharing a personal AI agent folder with team members causes the AI to speak on behalf of others in your name, and recommends instead providing read-only file access — via SharePoint, Google Docs, or MCP connections — so team members can pull consistent business context into their own separate personal AI systems.
- Paco argues that defining concepts with precise, unambiguous language is the critical foundation for effective AI use, claiming that if you deliver a clear three-line definition of a term like 'note' to an LLM, it will reference that concept correctly every time — and that this is exactly what IICOR's methodology provides as a structural advantage over improvised AI prompting.
Topics
Full transcript available for MurmurCast members
Sign Up to Access