I think I finally know what I want from all of this AI madness.
I want my own personal agent that I can collab with for years. Not just a chatbot that helps me write or code, and not some SaaS product that feels incredible right up until the company behind it changes the rules of engagement.
I want a system that actually feels like an extension of me. A second brain in the real sense. A place where my notes, morning pages, memos, goals, projects, links, drafts, screenshots, half-finished thoughts, and recurring questions can live and compound over time. And if that’s what I’m building, I want full control over it.
That part feels obvious to me now in a way it didn’t before. For a while, I was willing to trade control for magic. If the model was good enough, useful enough, human enough, I could ignore the rest. I could tolerate the outages, the weird product choices, the clunky interfaces, the feeling that none of this was actually mine.
But eventually that starts to wear on you.
Especially when you’re trying to build something deeper than a workflow hack.
The more personal software gets, the less acceptable it feels to have that relationship mediated by a company that can shut you out, rate-limit you, or decide your use case no longer fits their strategy.
I don’t want the thing that knows my writing, my taste, my obsessions, and my private thoughts controlled by any company. I don’t want to build a real memory layer on top of servers I don’t control using a model I can lose access to six months from now because demand spiked or incentives changed. If this thing is going to become part of how I think, it can’t be fragile. And it definitely can’t belong to them.
Closed models still feel more alive than anything else. That’s real. Claude Opus 4.6 feels closer to a legit collaborator than anything I’ve ever used. The conversations are warm. Fluid. Weirdly human. However, relying solely on OpenAI and Anthropic feels like a massive long-term risk.
I want hardware I can control, and ideally, models I can control and run on my own servers. I want a memory store I can inspect and back up. I want the vault to be mine. I want the agent to be mine. If one company disappears or changes direction, I still want the system. To achieve this, the primary tools one uses must be open.
Never bet against open source.
I still reserve the right to get seduced by some god model from Anthropic later. But if I’m serious about building a personal agent that truly feels like an extension of myself, then I need to treat it like infrastructure.
And infrastructure should be owned, not rented.
— Daniel