There is a quiet pattern in every AI engagement we've successfully shipped: the eval harness was the first thing we built.
Engineering
Evals before features: how we ship LLM systems
Our internal playbook for shipping AI features safely — and why eval harnesses are the first PR, not the last.
Imran Ali · Principal Engineer · February 19, 2026 · 9 min read
More posts
Engineering
The cost of mocking the database
Mocked tests pass while production migrations break. Here's a pattern we use to keep tests fast without losing the truth of a real database.
Apr 18, 2026 · 7 min read
Engineering
We stopped shipping feature flags in PRs
How a small process change cut our gate-cleanup debt to zero, with a one-line agent that opens a removal PR two weeks after a flag goes 100%.
Mar 22, 2026 · 5 min read
Design
Tile rhythm: a quiet pattern for marketing pages
Why we alternate light and dark full-bleed tiles on every site we build — and the rules that keep it from looking generic.
Apr 9, 2026 · 6 min read