Build in public · weekly

What we think about, in writing

Essays on AI-first teams, anti-gaming evaluation, and how RUQA is built. New post every Friday. No AI fluff — every word written by humans, then evaluated by humans.

2026-05-02Essay6 min

Stop tracking tools. Track meta-skills.

Cursor, Claude, ChatGPT, v0, Bolt, Replit Agent — by the time you finish your tools-used checklist, two of them are deprecated. The skill that doesn't go stale is the one that uses any of them.

Read essay

2026-04-25Essay9 min

Five signals beat one rubric.

Most performance systems collapse to a single score. That's exactly why people game them. We chose triangulation: five independent signals that agree or disagree on the record.

Read essay

2026-04-18Policy5 min

Why the first 90 days don't count.

If your team thinks the AI eval will be used against them in week 1, they'll game it from day 1. So we wrote a calibration period into the product, not the policy doc.

Read essay

2026-04-11Engineering8 min

Open-sourcing the judge.

When the algorithm decides who gets a raise, the algorithm has to be inspectable. Here's why we picked AGPL — and the small print on the commercial license.

Read essay

2026-03-28Build log11 min

March build log: sandbox v1, capability heatmap, and a bug we still don't understand.

What shipped, what slipped, what surprised us. One unresolved bug at the bottom — if you've seen it, please email.

Read essay

2026-03-21Essay7 min

AI-first ≠ AI-using.

Most teams use AI. AI-first teams have AI in the critical path. The difference shows up in standups, in PR reviews, and in how you measure capability.

Read essay

Get the weekly digest

One email Friday. New essay + one thing we learned. Unsubscribe in one click.