Claude finished the work. Then I lost it in the chat.
A creator demonstrates a comprehensive local-folder-based system for managing AI-generated deliverables asynchronously, featuring an interactive browser interface that tracks decisions, changes, and conversations without relying on external databases or web applications. The system enables collaborative editing, visual annotations, and automatic context preservation across multiple parallel AI projects.
Summary
The speaker presents a year-long iterative development of a workflow system designed to solve the core problem of losing control and context when working with AI across multiple projects. Rather than remaining tethered to continuous chat sessions, the system uses a local folder structure as a 'single source of truth' where all AI outputs, decisions, and conversations are stored and tracked.
The fundamental architecture leverages a folder-based approach with specific subfolders: a deliverables folder containing AI-generated work (images, reports, diagrams), a team inbox for new tasks, and structured markdown files documenting decisions and context. This structure works with any AI model (Claude, Gemini, etc.) because it relies on local files rather than proprietary platforms.
The interface built on top of this system provides a visual, browser-based dashboard that displays all deliverables with thumbnail previews, decision markers, and status indicators. Users can hover over items to see additional information, including how many open decisions require attention. The key innovation is enabling asynchronous work: users can review outputs, make decisions, add annotations (including drawings on images and timestamp-specific comments on videos), and create follow-up notes without maintaining active chat sessions.
The system tracks comprehensive audit trails showing which agents/sessions worked on each deliverable, what changes were made, and when. It automatically updates in real-time as new work is completed. Users can enter 'focus mode' on individual deliverables to handle one project at a time with full context, and can call the AI orchestrator (named 'Larry') for clarification without disrupting the workflow. The speaker emphasizes this represents true asynchronous collaboration: AI can pick up work from recorded context (including audio from calls) without losing progress if sessions terminate.
Behind the scenes, the system is remarkably simple—just an app created through conversation with AI that reads from local JSON files and markdown documents in the folder structure, serving the data to a frontend interface. The speaker built this entire system by talking to AI while iterating on design decisions, with styling and branding guidelines developed over months of parallel work. The solution maintains complete user independence from external services, requiring only folder backups to prevent data loss.
Key Insights
- The speaker describes sitting in chat sessions waiting for AI to respond and then manually finding relevant outputs as a significant productivity drain, which motivated designing a system that eliminates the need to keep sessions active.
- The system uses a 'decision warning' visual prompt system with emojis and brief code snippets instead of scrolling through long-form text, allowing users to quickly identify what decisions are blocking AI progress without information overload.
- The speaker demonstrates that by keeping multiple terminal sessions open for days working on complex parallel projects, tracked through a single deliverables folder, they achieve better organization than traditional chat-based approaches which lose context across sessions.
- The system enables direct inline annotation on generated content (including drawing on images and timestamped comments on videos) which then automatically translate into actionable follow-up instructions for AI in subsequent sessions without manual explanation.
- The speaker argues that reaching the level of polish and autonomous improvement shown in their system is impossible with standard chatbot interfaces because the AI learns and refines its own processes based on accumulated work patterns, not from explicit instructions alone.
Topics
Transcript
[0:00] I showed you on this channel already so many things you can do with AI that is actually useful, and how we organize everything using just a local folder and Claude. And today, I will show you something I've been thinking about for, well, a whole year to how to best work with AI, so whatever it creates for me, I don't have to sit in a session to keep control of it, keep tracking what it's doing, and then follow up the chat, and then go through the chat and find the things, and all this is confusing. [0:30] And in order to compensate that and make it clearer if I work on different things in parallel, inside…
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.