Nobody benchmarks agents with context loaded
Real agents have your journals, your docs, your history. Nobody tests what happens.
Blog
Why agent safety testing is hard to sell
It's a vitamin today. Six forces turn it into a painkiller.
Evals and RL environments are the same thing
Two industries share the same infrastructure and pretend they don't.
Announcing Ocarina Labs
What we're building, why, and how.