I’ve been trying to simulate using Codex for the next year and what will change about my perspectives on software engineering as I transition from being a computer programmer to a harness engineer. There are so many, but here are a couple that have stuck with me: Software

MCP = cat’ing every git man page into context Progressive disclosure: git --help, git rebase --help

This is fundamentally what the embedded knowledge base in the repository is for: design docs, ADRs, architecture diagrams, QA specs, known debt, critical user journeys, product feature descriptions, etc etc

This is the way news.ycombinator.com/item?id=471585…

This is the way

seatbelt-sandboxer > Generate minimal macOS Seatbelt sandbox configurations for applications github.com/trailofbits/sk…

Heard someone at unpromptedcon frame the harness engineering my team has been doing as getting the whole universe of non-functional requirements into code. This is a great way to think about things at a systems level

You can just (get the model to) do things

This OpenAI blog is a must-read. Don't ask what your agents can do for you. Ask what you can do for your agent openai.com/index/harness-…

Tokens are cheap

As a codex influencer,

Had a blast talking about how we vibe the security guardrails we want when tokens are cheap and code is free at #unprompted today with Paul!

I wonder what the equivalent is of LLVM’s Alive2 is for harness engineering

It is kinda wild it took as long as it did to go from machines that produce machine code to machines that produce bytecode to machines that produce code code

I do often throw work away. Kinda like aggressively saving in GTA before a mission so you can rollback if things go sour

OH at #unprompted: malware to start embedding the Anthropic stop token all over the binary to tell Claude to stay away

This is my team h/t frantic who ships 75 PRs a week with Symphony. We shipped a spec. Come vibe your own