I spent an afternoon researching AI tools for personal life management — the kind that promise to be your digital chief of staff, your second brain, your always-on executive assistant. Dot, Mem, Lindy, Motion, Rewind. I went in expecting to feel behind.
The life AI market is real and growing, but oddly fragmented. Each product has picked a lane and defended it well. Motion reschedules your calendar when tasks slip. Lindy connects two hundred-plus apps into no-code automations. Mem organises your notes and surfaces connections you didn’t notice. Each is a polished answer to one question, and none of them talk to each other. More importantly, none of them know anything about your life that you haven’t explicitly handed them — which turns out to be most of it.
Then there’s the product that doesn’t exist anymore. Rewind promised to capture everything passively — every screen, every conversation — and make it searchable. That was a genuinely different idea, the kind that could have anchored a real personal AI OS. Meta acqui-hired the team in December 2025. The app shut down ninety days later. Build your cognitive infrastructure on someone else’s product, and you inherit their business risk.
By this point in the research, I was expecting to choose between narrow products and fragile ones. What I found instead was that the practitioners who write seriously about using AI for knowledge work have quietly converged on something different. Not a product. A stack: a language model with tool access, a local markdown vault, and custom scripts. A coding agent, repurposed. Which makes more sense than it sounds, because a life is, in some sense, a repository — a corpus of accumulated decisions, open threads, and learned patterns that you are constantly reading and occasionally updating. The tool designed to work with repositories turns out to be the right tool for that.
The “wrong tool for the job” framing inverts when you look at why it works. Code demands precision about state. Files are files, commands do what they say, output is inspectable. Every dedicated life AI product draws a circle around what it’s allowed to know — Notion’s agents can only read Notion, Lindy’s automations only reach its approved integrations. When the foundation is a shell and a filesystem, that circle doesn’t exist. The constraint that makes a coding agent uncomfortable for casual use — you have to be specific, you have to describe state, you have to name your tools — turns out to be the property you want in an operating system.
There’s a deeper point about auditable memory. Dot, the most relationship-focused of these tools, builds what its creators call a “theory of mind” portrait of you — routing your queries through multiple models filtered through accumulated context. The idea is appealing. But when it’s wrong about you, there’s no diff to read and no commit to revert. A vault of markdown files I’ve been editing for years is different in kind, not just degree. Every assumption the AI makes about my priorities is grounded in something I wrote and can overwrite. That’s not a technical advantage; it’s an epistemic one.
None of this is to dismiss the dedicated products. Motion’s dynamic scheduling is genuinely hard to match with scripts. Superhuman is real productivity gain if email is where your time goes. These are best-in-class tools for specific problems. The mistake is expecting any of them to be a platform.
What I built, accidentally, is a platform. The accident was starting with the wrong tool. I had been treating the coding constraint as a limitation to work around. The research afternoon clarified that it was the feature.
Tools solve problems. Infrastructure disappears. The coding tool, it turns out, was always pointing at that. I just needed the comparison to see it.
P.S. The one thing genuinely missing: passive capture. ScreenPipe — open source, local-only, MCP-integrated — fills exactly this gap and plugs directly into Claude Code. I haven’t added it yet, which means I’m still the manual layer between experience and memory. The stack is good. It’s not finished.