Category: Uncategorized

Nines are earned, not claimed. A look at the guardrails, tests, and on-call culture behind our uptime — and the postmortem habit that keeps us honest. We treat every incident as a gift: a specific, reproducible way the system surprised us. Fixing the class of bug, not just the instance, is what compounds into reliability.

Latency is the difference between a tool people reach for and one they avoid. Over two months we took our p95 from 900ms to under 350ms. Here is the engineering work behind it. Measure first We instrumented every hop and found the surprise most teams find: the model was not the bottleneck. Queueing, cold caches,…

A good changelog is marketing. Lead with the benefit, not the version number, and write like a human. This post breaks down the format we use for every release. Every entry answers one question: what can I do now that I could not do yesterday? If we cannot answer that, it is a patch note,…

Activation happens fast or not at all. We rebuilt onboarding around a single first win — one thing a new user can do in under five minutes that proves the product works. Everything else — settings, integrations, invites — waits. The job of the first screen is one successful action, not a tour.

Our v2 model is twice as fast at the same quality, and it costs half as much per token. Here is what changed under the hood, what it means for your product, and how to upgrade in a single line. What changed We rebuilt the serving stack around a smaller, distilled model and a smarter…