@matryer What does your workflow look like when building a new feature or debugging an issue?
How do you structure prompts, and do you maintain specs in your repo if so, what do they look like?
@thorstenball Would love to know how you organize an avg day like... It feels like you have more time per day than the rest of us, I for one would benefit from knowing how to learn from your routine and thinking process
Shipped a tiny offline word party game on Android 🎉
https://t.co/fpZSQj0WCJ
No accounts, no ads just pass-and-play.
Mostly testing if the core idea is fun.
If you enjoy word games, feedback welcome.
If an agent can’t navigate a repo, understand what “done” means, or converge without constant correction, that’s not an agent failure.
It’s a repository and context design problem.
This is why teams are moving toward specification-driven development
With more AI agents writing and modifying code, a pattern is becoming obvious.
Many “execution problems” only exist because repositories are designed for humans to supply context, explain intent, and redirect work mid-flight.
Agent work needs to be disposable by default:
– cheap to create
– cheap to discard
– impossible to accidentally trust
Full essay here:
https://t.co/e69zo3xHuK
1/
The hardest part of building software in the AI era isn’t only coding.
It’s alignment.
Code is relatively easy.
Agreeing on what the code should mean is hard.
That’s why I’ve shifted to Spec-Driven Development (SDD). 🧵
8/
Big thanks to:
Spec Kit - https://t.co/iS8SVS2Kiq
Mastering AI Agents by Mario Cartsburg
Both shaped how I think about this. - https://t.co/11r5RbMVIE
🌀 noise. 🔥 fire. 🎶 tones.
Inspired by a prompt in @thorstenball’s newsletter, I built a terminal synth that animates ASCII, shifts color palettes, and plays real-time sound.
Deno + OpenSimplex + curiosity =
https://t.co/8FutFk73Ax
@AnthropicAI@deno_land