Featured · Guides
Read the post
Progressive delivery for LLM prompts
Progressive delivery for LLM prompts
Treat a prompt like a deploy: version it, roll it out by percentage and roll it back on a judge score — not a hunch.
Notes from the release train.
Engineering
How we evaluate a flag in 40 ns
Inside the Go SDK’s zero-allocation hot path and the benchmark that keeps it honest.
Agents
Approval gates without the bottleneck
Which agent actions deserve a human, and how to keep the queue under a minute.
Customers
Kilnworks retired its release freeze
Nine hundred services, zero freeze windows, and the rule that made it safe.
Experiments
Sequential tests you can peek at
Why fixed-horizon A/B tests break when you check them daily, and what we do instead.
Releases
What shipped in August
Parquet export, the MCP server, judge-aware rollbacks and forty smaller things.
Security
Policies as code, reviewed like code
How to put agent guardrails through the same pull request as the rest of your stack.