I never could have predicted this. Last year when I posted the list of software projects I’d built it felt like that was somewhere close to peak AI software. Like I knew it’d get better but I never expected it to get this good. Over the last few days, since Opus 5.5 specifically, a model that is exactly 1 version away from the model that first blew my mind with its capabilities about this time last year, I’m not really sure what to do about software. An X post by DHH I saw today said something like “all single-user software has lost its moat.”
I’d extend that to “all software that one person can plausibly maintain has lost its moat.” That excludes a lot of software, notably most enterprise software or mission-critical software for most businesses, but it sure does include a LOT. Most desktop apps, most games, most editing software of all types, most websites, most software that runs on unlockable hardware, etc. etc. That’s a lot of power. It’s also a sort of weird place to be.
I don’t think I like writing software to feel special. I do like writing software to share cool things with people though and … I’d like to write software to make money. Now, pretty much anything I can imagine can be built over anywhere from a few minutes to a few days. It’s hard to fathom. I’ve built SO MANY THINGS. A podcast that’s got well over 100 episodes in both audio and video. A infinite world driving game with all kinds of scenery and driving modes, tons of developer software, and… really just too much to even begin listing. I’m pretty confident that, at this point, pretty much whatever I dream up can be built and, with Opus 5.5, built probably better than I can imagine it.
That’s… insane. I can hardly believe it. I can also hardly believe that I can’t think of a ton more to make. The reason is maybe not what you’d think it’d be. It’s because either software already does everything I can think of or I know that if I want it to do something that it doesn’t do already that I can probably have an AI agent do it completely bespoke when I need it done. It’s like I bought the candy store. I can only eat so much candy now that it’s all mine.
I’m sure this is just another weird phase in the many weird phases over the last year. Maybe I’ll figure out how to think bigger or how to get more interested in the set of problems that aren’t fully solved. As far as I can tell, they’re the ones that need corporations and teams to work on. Not so much for the software but for support, sales, integrations, etc. Either that or I’ll get into making stuff in the Real World. 3d printing, robotics, or who knows what else. All of these hobbies that at one time felt that they’d take a long time to get into and would, therefore, that I’d probably never do unless it was in another life. Now almost nothing is intimidating. I have, we have, access to what I’d be fine with calling a superintelligence. It’s a bit of a shame that Bostrom’s term, at least his book Superintelligence is where I first saw it, has been politicized but whatever, none of that changes the fact that we have a superintelligence, at least one, Astra’s pretty darn good too, at commodity prices.
It’s so easy to imagine a world where Opus 5.5 existed but was totally inaccessible except to Enterprise.
It’s amazing that this is the world we got.
It’s wild watching people wake up to it over this year. I think most people still haven’t really though. If they’re using ChatGPT to format documents and help them with relationship problems (both of which it’s better than most people at), then I imagine it’s hard to tell really how far things have come. I’ve seen a single agent work for days on end without intervention on projects that would have taken any dev I’ve ever met months, probably years to do. And it’s damn near free.
So… 2026 was the year of building any software I could imagine. It seems it was that for me and for a lot of other people. I’m not sure what 2027 will be. Will it be the year of bespoke everything? Will it be the year that software begins to enable experiences in the real world that were previously unattainable or even unimaginable? It sounds cracked but… at the beginning of this year I thought it was amazing that Claude could help me redo my blog or make dev tools. Now a lot of what took me days or weeks at the beginning of the year I can essentially one-shot. The only limit is my ability to describe it and give the agent the right environment.
It’s wild times. On one hand, life going on as it always did, on the other hand, and I suppose you could say this at most any time in the last few thousand years, life will never be the same again. This time though, it’s hard to even come up with comparisons. A lot of people still seem to see AI as something they chat with to get their job done a little easier. It is that and it’s about to be so much more.