Engineering
The best engineering practices from the team building the future of AI.
We spent six months trying to make AI coding tools work. The models made changes without clear criteria. Development velocity didn't improve. The problem wasn't the AI. The problem was our code.
Most teams using AI for coding jump straight to implementation. For brownfield codebases, this breaks down. We developed a four-step workflow that solves this: Spec, Research, Plan, Implement.
Every team using AI coding tools hits the same wall: the model writes correct code that violates your architecture. The fix isn't better prompting. It's a documentation layer that gives AI agents persistent context.
When AI handles the mechanical coding, the bottleneck shifts. The hard part is no longer writing code fast enough. It's understanding the problem well enough to direct the AI correctly.
Most teams answer 'is AI working?' with vibes. We use a single metric: the percentage of reported bugs that can be resolved with a single prompt. No vibes required.
There's a spectrum of AI coding assistance. On one end, autocomplete. On the other, full autonomy: the model takes a product requirement and ships a working feature. We've been pushing toward the autonomous end.
Most content about AI in companies focuses on engineering. If you only use AI in engineering, you're leaving massive leverage on the table. Here's what we use, where, and why.