I’m a developer of 30 years, but writing a game completely with AI. It writes all the code and has something like 200 test cases now . It does well with what it understands but there are all kinds of things it fails at. Like it didn’t blink at an interactive turn-based game taking 10 minutes per turn (instead if an expected 15-30 seconds). That was caused by recomputing everything rather than caching, which I pointed out. There were 3 areas that referred to something we changed. I had to point out what was no longer coherent. AI can write code well, but it just doesn’t have the memory/context size to drive a real world project of any size. It still needs someone else to do the thinking and planning. If I didn’t code I don’t think it would have known where to look to fix many performance problems. I’m not at all worried about it taking my job.
some/most of that could easily be optimized for: set expectations for turn-time ahead of time, and it will meet them. if you had even just given it a clock to pay attention to, it would have noticed on its own, i reckon.
i'm not having problems yet in my biggest ai-only codebase, which is over 100k lines now. i haven't had any model hiccup, actually, but the smaller models take more wall time to get it done.
i have the models writing in rust, i believe it helps. ymmv.
i don't think it always architects well or anything, but it has quit building broken shit. it gets lazy, builds dumb UI, puts way too much text everywhere--but only if i don't remember to give it some better guidelines, or a visual reference.
i think with the right context, they build the right thing, and we haven't totally solved that just yet. but close enough for me; i'm so thrilled. i just want opus 5.5 to be crazy fast, like chatjimmy.
Well, my project IS complete, although I keep adding new stuff--it's going to be that kind of app, it rotates content in (minigames).
It has about 8 minigames, leaderboards, awards, achievements, async image generation, inventory, inventory-interaction-with-minigames, and various other account features, mailers, notifications, blah blah.
Literally haven't had one issue with context or micro-managing the agents on this project, iirc. I built it in a couple weeks.
Minigames are like "putt putt golf w/ daily generated levels", "slot machine," "video poker", etc. Not like "tap the fastest" or something with no complexity.
ks2048 · · focus · HN ↗
With the right high-level abstractions, I can write code as fast as trying to convert that thought into English (for some tasks).
sejje · · focus · HN ↗
How fast can you scaffold new ideas and delete them when they don't work?
DougN7 · · focus · HN ↗
sejje · · focus · HN ↗
i'm not having problems yet in my biggest ai-only codebase, which is over 100k lines now. i haven't had any model hiccup, actually, but the smaller models take more wall time to get it done.
i have the models writing in rust, i believe it helps. ymmv.
i don't think it always architects well or anything, but it has quit building broken shit. it gets lazy, builds dumb UI, puts way too much text everywhere--but only if i don't remember to give it some better guidelines, or a visual reference.
i think with the right context, they build the right thing, and we haven't totally solved that just yet. but close enough for me; i'm so thrilled. i just want opus 5.5 to be crazy fast, like chatjimmy.
gridspy · · focus · HN ↗
But due to the sunk cost fallacy and the belief that others have made this work you'll keep plowing money into the AI Slot machine.
sejje · · focus · HN ↗
It has about 8 minigames, leaderboards, awards, achievements, async image generation, inventory, inventory-interaction-with-minigames, and various other account features, mailers, notifications, blah blah.
Literally haven't had one issue with context or micro-managing the agents on this project, iirc. I built it in a couple weeks.
Minigames are like "putt putt golf w/ daily generated levels", "slot machine," "video poker", etc. Not like "tap the fastest" or something with no complexity.