I've been writing code for as long as I can remember. There's been a lot of debate about AI-generated code and whether anyone even reads it anymore. A few thoughts on coding agents, engineering responsibilities, and why nobody reads all the code anyway.

Claude Code feels like the engineer who overthinks the solution and then writes a long exposé on how clever it is. Codex feels more like the quiet engineer who just gets on with it.

Every new software project begins with the comforting illusion that this one will be relatively straightforward.

It would appear as if Claude Code has finally become a true engineer.

I found this doodle I made in one of my countless notebooks, and turned it into a 3D character. Kind of like, feels like it could work for a game. Needs a cool name though, suggestions?

Now, that’s a car reveal. Insane.

What's the verdict on X Money now that it's been out for a while? Anyone using it? What are the biggest downsides? X Premium is a no-brainer for my use case, so I'm tempted to give X Money a try as well.

There's a lot to be said for building things simply because you want them to exist.

Making something simple is often considerably more difficult than making something complicated.

Other than ordering groceries, what do people actually use personal AI assistants for?

Small but important update to Otone before the first App Store submission: external audio interface support on iPhone/iPad. Sometimes you just want to record some quick ideas without having to start up the computer.

At some point, it feels like productivity becomes procrastination with better marketing.

There's more value in knowing what to build than how to build it.

4:50 PM on a Friday. “Hey, do you have 2 minutes for a quick question?”

Preparing Otone for iOS and macOS App Store submission. What are some common, easy-to-overlook reasons for rejection?

After a bit of hiatus I wanted to get back into some graphics, so naturally I decided to revise my C++ reflection system instead. I'm OK with the feature set, but I feel the syntax is too verbose. I don't like using macros. Any suggestions for a clean reflection API?

For someone who's used to writing C++ with minimal external dependencies, seeing GPUI bring in about 426 dependencies is somewhat of a hard pill to swallow.

With that said, running MTPLX with Qwen3.8-27B-MTPLX-Optimized-Speed provides a significant speedup.

Started to build a local AI agent orchestrator / control plane. I wanted Split View for multiple live project sessions in one window. No more jumping between terminals/chats. Early + buggy, but already quite useful.

What's the current consensus on Grok Build? Does it still upload the entire repo to a server?