@ezyang Better automated testing suites and proper target determination, clear coding rules and standards as well as project vision
I also think it requires a big up front cost to set these up.
But once you set up a project with these things I think it becomes a lot easier to iterate.
I feel like most agentic interfaces are operating at a level like a micromanaging front line manager.
There needs to be an evolution where we can set up agents to do most of the work without needing to be so in the loop.
One of the biggest strengths of GitHub actions is the lack of needing to do an extra step to setup a workflow
If I commit a yaml file in the .github/workflows directory it does the thing and people underestimate the power of that.
The fact that you have to use windows with this machine just makes it a non starter for me.
The last thing I want is ads in my OS for a $3000+ machine.
Introducing Surface Laptop Ultra.
Built for world makers. Designed for what's next.
The most powerful Surface laptop ever. Coming Fall 2026.
Sign up to learn more: https://t.co/k8aEX2pTAy
Fan control is such an nice quality of life piece of software.
My computer no longer sounds like it's about to take off when I'm just watching a youtube video.
My manager after I explain to him that my one opus 4.8 ultra code prompt just used up our Claude budget for the entire month.
The git commit wasn’t gonna write itself fwiw.
I feel like a lot of time being spent building harnesses right now is focused on this core dev loop as if you’re a front line engineer.
As engineers get better at automating this dev loop we’ll naturally evolve to higher level concepts in harnesses like project management.
@difficultyang@drisspg I usually use pi as a direct replacement for codex + Claude code. When I was still using opus I found pi to be a much better harness for opus directly.
The nicest thing about pi is if the harness doesn’t do a thing that you need it to do it’s so easy to extend it.
@charliermarsh This reminds me of when PyTorch had to create its own s3 artifact solution since GitHub’s upstream artifacting solution couldn’t handle a 1GB artifact without hitting a 503.
A little secret. About 5% of our production traffic is on the Pi harness, about another 5% is on OpenCode. Reminder you can use your ChatGPT account in a flourishing set of other tools.
We’ll continue to make Codex awesome, but you have options.