Tom Solid | AI Productivity @TomSolidPM
AI doesn't fix broken systems. It amplifies them. I help professionals build the system first. Creator of ICOR® myicor.com Joined December 2010-
Tweets8K
-
Followers4K
-
Following167
-
Likes6K
See the full breakdown of my new One Folder + Claude approach here 👇 youtu.be/51ZVWAjHurI
Notion. Tana. Heptabase. Obsidian. Day One. I moved my knowledge across all of them over the years. Every move, the notes survived the export. The links between them did not. That is the part nobody warns you about. Your second brain is a hostage. The tool owns the format, the tool owns the connections, and when you leave you leave most of the value behind. The value was never the notes. It was the links. And the links never make it through the door. Which tool did you lose the most in when you migrated, and did the links survive?
A chat window is a stranger every morning. You open Claude or ChatGPT, and it knows nothing about your business, your people, your decisions, or how you write. The auto-memory they sell you is thin. So you spend the first ten minutes re-explaining yourself, again. A folder is the opposite of a stranger. A folder remembers. I point the model at a folder of plain text that already holds all of it, and the context is just there before I type a word. The strange result: my prompts got shorter over time, not longer.
Do not copy my folder. I mean that. The reason templates fail is the same reason every tool migration failed me. A structure built for someone else's life does not fit yours, and forcing it just adds noise to a system that was supposed to reduce it. I run zero downloaded plugins and zero community skills. Not because they are bad. Because the less I import from other people, the more the folder fits the one life it is actually for. Start with the smallest version. An orchestrator, a researcher, someone to file what you capture. Then grow it one role at a time, written for you. The method is portable. The specifics have to be yours. What is the one template you downloaded with great intentions and never actually used?
The companies failing at AI are failing the same way they failed at automation for twenty years. They drop a new tool on a broken workflow and wait for the workflow to fix itself. I watched it happen inside real companies for most of my career. New buzzword, same broken process. The team goes to coaching, comes back, and works the same way with new vocabulary. AI does not break that pattern. It accelerates whatever you point it at. Point it at a workflow you do not understand and you get a faster way to stay confused. The unglamorous work comes first. What do you actually do all day. What are the steps. Where is the friction. Write that down. Then AI can accelerate it. It does not work in reverse. My folder works because I understood my own work before I handed any of it to a model. The expertise stayed mine. The friction is what I gave away. Have you mapped what you actually do all day, or are you adding tools on top of a process you never wrote down?
Most PKM systems collapse because people invent a new bucket every time something feels different. The shape is simpler. Four buckets. Four questions.
People ask why I do not just use ChatGPT or a Claude chat window for this. The answer is one word. Context. Every chat session starts from zero. The auto-memory they sell you is thin. It does not know my business, my people, my decisions, or how I write. I run everything from one folder of plain text instead. CRM, projects, journal, research, all of it. The model reads the folder, so the context is already there. I never paste it into a prompt. The strange result: my prompts got shorter over time, not longer. When the structure does the remembering, prompt engineering barely matters. How much of your day goes to re-explaining context to a tool that forgot it again?
@ClaudeDevs Good to see Claude starting to natively implement verification feedback loops. Something I’ve had running for a year now, and all based on a local folder structure. x.com/TomSolidPM/sta…
Most AI setups are one assistant trying to do every job. I run a team of specialists instead. I talk to one orchestrator, Larry. I never talk to the rest. I hand him a request, he picks the right specialist, hands off the work, a QA pass verifies it, then he brings the result
Good to see Claude starting to natively implement verification feedback loops. Something I’ve had running for a year now, all based on a local folder structure 👇 x.com/TomSolidPM/sta…
How do you get Claude Code to check its own work before handing it back? Watch how you can encode your manual checks so Claude closes its own feedback loop:
@itsolelehmann I switched off Auto Memory in Claude because using persistent memory within just one local folder is much more reliable and token-efficient. This worked long before Anthropic released any memory-related updates, and it still outperforms them. x.com/TomSolidPM/sta…
In the video I run a live test. Same prompt, same model. One run inside my folder, one inside an empty one. The empty folder does generic web research and hands back a generic infographic and a flat post. My folder routes through the team, asks scoping questions, pulls from my
I show all of it on screen in the full walkthrough. The real folder, the empty-folder comparison, the team running in parallel, and how the whole thing backs up to a single drive with no cloud bill. If you want the structure, it is free. Watch the walkthrough first: youtu.be/51ZVWAjHurI
In the video I run a live test. Same prompt, same model. One run inside my folder, one inside an empty one. The empty folder does generic web research and hands back a generic infographic and a flat post. My folder routes through the team, asks scoping questions, pulls from my own knowledge, and matches my brand and my voice. Same model. The folder is the difference.
One folder of plain text on my machine runs my whole life and business. CRM, projects, journal, research, all of it. Claude reads the folder and works inside it. No app subscriptions. No cloud. No context lost between sessions. Here is everything in the walkthrough, in one thread.
@venturetwins Me asking Opus 4.8 what time it is
You might like
