✦ Agentic AI Developer ✦ Sr. Software Engineer ✦ Got Idea? Let’s Make Free MVP
Joined September 2024
- Tweets405
- Following18
- Followers10
- Likes700
Show me what you build, most of ppl don't care about, but I will check, visit, and send you feedback, and could be your potential pro user.
It’s very new trend playing in your browsers, it reminds me of Counter Strike game.
Love it men @levelsio
I keep going more niche with reverse engineered web-based versions of games
👉 ut.pieter.com 👈
6 months ago I tried to reverse engineer Urban Terror
Urban Terror was kind of a Counter-Strike style mod on top of the Quake 3 Arena engine, I think @Erwin_AI got me to play this during COVID
Because it runs on Quake 3's engine like RTCW and Q3 itself, I thought I could just use ioquake3 and WASM to run it on web, then add a multiplayer server and it'd just work...
But no it didn't, I got stuck 6 months ago, it was buggy and wouldn't work, I kept trying and kinda got it to load but not really play
Today I tried it again with Opus 5.5 and again, I couldn't get it to work, giving up I did a last try with Fable 5.1 on high effort and it after awhile it did figure out the issue!
"The engine's struct was 4 bytes longer than what UT's bytecode expects, everything after that point (the entity count and the entity list) was read 4 bytes off"
It rewrote the bytecode and made it work!
So now we have another game preserved, running 100% the web via WASM, with a multiplayer server tied to it, so you can play with your friends 😊
Every model that comes out makes a bit more things possible and you can just keep trying old things that didn't work before
I see lots of people mod modern games now too which looks very cool too
Here I'm playing it with @marckohlbrugge a few minutes ago! Anyway you can play it with everyone else at ut.pieter.com now :D
What if we created great animations with strong stories using tools like Opus5.5 ? Would it be worth it?
There are repos now with a README for humans and an AGENTS.md for AI.
That still feels slightly sci-fi to me.
We’re not just writing software with AI anymore.
We’re starting to design codebases that AI can actually work inside.
God knows what's next?
One thing I’m starting to hate in AI products:
choosing the mode before I even start.
Chat. Agent. Research. Code. Canvas.
I don’t want to learn your product’s org chart.
I want to say what I need and let the system figure out how to do it.
The IDE is quietly becoming a control room.
One agent fixes a bug.
Another writes tests.
Another works on a feature.
The weird part?
I’m spending less time asking “how do I code this?”
and more time deciding what should happen next.
Funny how detailed tickets used to feel like bureaucracy.
Now I give the same task to an AI agent and suddenly every vague sentence comes back to punish me.
The better these tools get, the more obvious one thing becomes:
clear thinking was always part of the code.
AI has ruined my patience for software.
If I open an app and have to hunt through 6 screens to do one obvious thing, my first thought is now:
why am I still doing this myself?
What app feels painfully pre-AI to you?
I’m not sure if I’m learning… or just borrowing intelligence.
I caught myself asking AI for an answer before I’d even formed my own opinion.
It was faster.
It also felt a little wrong.
So lately I try to have a guess first even a bad one
I don’t care that much if AI saves me 30 minutes writing code.
What excites me is this:
one engineer can now attempt things that would've needed a small team a few years ago.
That’s the shift I’m interested in.
Not faster coding.
Bigger ambition.
I didn’t expect AI to make me miss friction.
Before, an idea had to survive enough effort to earn a prototype.
Now I can build it before deciding if I even care about it.
That’s powerful.
And honestly, a little dangerous.
There’s a weird guilt in software work!
if I’m not typing, I feel like I’m not working.
Code is visible. Thinking isn’t.
So it’s easy to spend 2 hours “making progress” on something 20 minutes of proper thinking could’ve killed completely.
I’m trying to get better at that.
The strange part of coding agents for me isn’t the speed.
It’s looking at several tasks moving at once and realizing I’m no longer doing all the work myself.
That feels powerful.
And slightly uncomfortable.
Because now the real work is deciding what deserves my attention.
Building is getting ridiculously cheap.
And that excites me… but also makes me a little uneasy.
When almost any idea can become a product in a weekend, I don’t worry as much about whether I can build it.
I worry about whether it deserves to exist.
AI can generate code faster than I can properly understand it.
That creates a new kind of technical debt:
understanding debt.
The code works.
The tests pass.
The PR ships.
But nobody really owns the mental model.
I think this will become a much bigger problem than we expect.