Why your service mesh is probably wrong
Most teams adopt a service mesh because it sounds like the right thing to do. Then they spend six months debugging mTLS cert rotation at 3 a.m. Here's how we think about it instead.
BLOG
Engineering writing
Technical writing from the studio. No marketing fluff. Just the things we've learned building systems that run in production.
The real cost of real-time
Everyone wants sub-second dashboards until they see the infrastructure bill. We break down when streaming makes sense, when batch is fine, and how to build a cost model that doesn't surprise you.
Secrets management without the drama
Vault, AWS Secrets Manager, environment variables, encrypted files — every team does it differently. We've migrated secrets architecture for seven clients. Here's what actually works.
Evaluating LLM outputs in production
Most AI deployments have no evaluation. The model works in a demo, ships to production, and drifts silently. We describe our eval harness: golden ticket sets, automated regression detection, and a feedback loop that catches quality drops before customers do.
Temporal in production: what we've learned
We've run Temporal in production for three years across five clients. It's powerful, but it has edges. Here's our playbook for workflow design, failure handling, and operational hygiene.
On-call for systems you didn't build
Sometimes you inherit a system that has no runbook, no monitoring, and a bus factor of one. Here's our 72-hour protocol for taking ownership of an unknown system without breaking it.