Using AI

What follows describes how we actually use AI today. We take a descriptive, not prescriptive, approach: this is a snapshot of what’s working rather than a rulebook, and it will keep changing as we do.

Culture

Our culture around AI is simple: use it, push on it, and share what you find. We’d rather someone try something that flops than wait for permission.

  • Reach for it. AI is the default, not something you ask permission for.
  • Push on it. We try new tools and workflows constantly, and the dead ends are how we learn.
  • Own your workflow. No single right way. Build what makes you fastest, and change it when you find better.
  • Share the wins. Found a prompt or skill that rips? Tell everyone.
  • Spread it company wide. Not just an engineering thing. We run regular trainings to help other teams put AI to work.
  • Backed from the top. Leadership encourages big swings over incremental wins, and puts real time and money behind them.

How We Work Today

These days we’re rarely writing code by hand. Most of it runs through Claude while we steer, review, and refine. The specifics shift constantly, and some are still experiments.

What We’ve Used

  • Claude Code
  • Codex
  • Conductor
  • Compound Engineering
  • Dedicated Slack channels for sharing what’s working
  • Skills any engineer can invoke: opening PRs, QA testing, fixing specs.
  • Agent contexts (CLAUDE.md files) seeded through the codebase, sharpened as we learn what helps.
  • MCP servers that connect the tools to the systems we use.

Ownership

We move fast with AI, and we own everything that comes out of it. That ownership is what keeps quality high, and the results have followed.

  • It’s ours to own. However a change gets written, whoever ships it stands behind it.
  • AI reviews everything. Every change gets an AI review before it lands.
  • Tests and linters back it up. A comprehensive test suite and opinionated linters catch regressions before they ship.
  • Humans for the big stuff. Significant change or a shaky approach? We pull people in.
  • The numbers back it. Same size team, roughly double the deploys. Monthly deploys per engineer went from around 20 in 2024 to the mid 30s in 2026, hitting an all time high of 44 in June 2026.

Further Reading

A few pieces that have shaped how we think about and work with AI.