Telos — Declarative software, managed for you
usetelos.ai · 1,402 words · saved by 1 readers
Long-running coding agents need more than persistence. Adversarial verification is the feedback loop that lets software begin to maintain itself.
I've started to think of programming lately as holding AI on a leash. Pull too hard and you may as well write the code yourself. Let go entirely and you're drowning in sloppy code that you don't meaningfully understand. I can now write code faster than ever, but that simply creates... a whole lot more code to review. As an engineer, my role has become a dance between prompting furiously and deleting carefully to not lose coherence of my software. In between defining my intent and thinking about system design, I mostly feel like a human gatekeeper. How do you define slop? I recently came…
saved by
related reading
- Your job is to deliver code you have proven to worksimonwillison.net
- If coding is solved, what now?: Measuring the sloppiness of code | EARENDILearendil.com
- Demystifying evals for AI agents \ Anthropicanthropic.com
- Designing agentic loopssimonwillison.net
- Thoughts on slowing the fuck downmariozechner.at
- claude code is not making your product betterethanding.substack.com
- Effective harnesses for long-running agents \ Anthropicanthropic.com
- Harness design for long-running application developmentanthropic.com
- Coding Models Are Doing Too Much | whnrehiew.github.io
- Agentic Coding is a Trap | Lars Fayelarsfaye.com
- Vibe engineeringsimonwillison.net
- When AI Writes the World's Software, Who Verifies It? — Leonardo de Mouraleodemoura.github.io