Felipe Arce

whitepapers.exowatt.com Moving 199 repos off GitHub Actions in eight days — Exowatt Research After a three-hour GitHub Actions outage, we built Roadrunner, a self-hosted Woodpecker and Dagger deploy lane. Eight days later it was the default deploy path for 199 repos, with GitHub Actions kept...

Exowatt @ExowattEnergy · 2d

Our software team has been doing amazing work this year

Physical AI is at the core of what we do, and the tools we built for it are too good to keep to ourselves

So today we're launching Broad Daylight, our new open source initiative. Read about it here:
whitepapers.exowatt.com/broad-daylight

View quoted post on X

Allan Verified account @AllanatrixQ · 2d

I’ve been super pro open-source models and all, but honestly, the frontier models still beat the open ones in quality. I’ve tried dozens of other models and fine-tuned them and post-trained the crap out of them, but for the kind of work I do and the heavy research infrastructure and ML work, frontier models still win. The open models are great for narrow tasks that don’t require too much open-ended reasoning, but not so much large-context tasks, and multi-step and branching work; otherwise, open models are sufficient.

View quoted post on X

whitepapers.exowatt.com Feads, a work board for agents that people read — Exowatt Research We turned beads, an open-source issue tracker for coding agents, into Feads: one hosted board where nine people's agents claim work, declare the files they touch and @mention each other. Since July...

Logic Verified account @code0e · 2d

For developers using Claude Code and Codex together, how do you prevent both agents from making conflicting changes?

Do you assign separate tasks, use different Git branches or give each agent its own worktree?

What’s actually working?

View quoted post on X