Software Finally Starts Adapting to Me
I've had a very strong feeling lately: I increasingly don't want to learn software anymore.
It's not just laziness — though I am lazy. More fundamentally, the old software logic was: you adapt to me. Where the buttons are, how the menus hide, how the workflows twist — you have to learn it all. If you can't learn, you're stupid; if you can't remember, you're old. Software features multiply, menus grow ever more complex — 90% of which you'll never use in your lifetime — but vendors can't restrain themselves from expanding coverage. This is a kind of "collective menu debt," yet every individual who only needs a fraction of those features must still repay it, must learn to penetrate the complex UI to find their own subset.
But now that AI agents have arrived, this logic can be reversed. A friend who develops agent platforms advocates exactly this, saying conditions are ripe to build software just for yourself.
In fact, I've recently been using Codex to build a tool specifically targeting my own pain points from years of digital life: an automated system that collects anything I'm interested in, auto-classifies, processes, structures, and archives it, ready for retrieval and summarization at any time. I don't need to learn it, because it grew out of my own habits. The ideal state isn't me adapting to generic software — it's custom software adapting to me.
This kind of software has one enormous advantage: it has no market, therefore no competition. It serves just one person. It doesn't need to please investors, chase DAU, pursue growth, or design "user retention." It just needs to make my life smoother, help me lose fewer things, help me think more clearly, and automate the manual workflows I used to do. That's enough.
Which brings me to a regret.
Looking back on my life, my deepest source of inadequacy is that I didn't study science or engineering as an undergraduate — I studied humanities instead. (It really wasn't my fault — I applied for science and engineering, but the first cohort of post-Cultural Revolution college entrants in 1977 barely knew English, so English wasn't a required subject but could be taken as a bonus. I thought the bonus English test would help my application, but the foreign language department, desperate for English-capable students, forcibly pulled me in. No negotiation.) But your first degree is, in some sense, your underlying operating system. If your foundation isn't solid enough, you can patch it later, upgrade it, install plugins — but that gap in fundamentals will always be there. This has been my Achilles' heel for decades.
Fortunately, large model agents have arrived. My requirement for myself is now simple: since I didn't study enough before, let the tools fill the gap. Let coding agents become my private science-and-engineering assistant and personal secretary. They don't replace my judgment, but they compensate for my weaknesses. I don't need a market-facing software matrix. I just need an increasingly handy, increasingly understanding toolbox.
Efficiency first, fit first. If it can help me retain what's in my mind and bridge what I didn't learn before, that's enough. This "personal dynamic knowledge base" agent is no simple project, but it's nearly operational. Looking at it now, building your own wheels for your own use isn't actually that hard.
