Thinking about agents
in production.

Why agent evals that pass in development fail in production

Most eval failures are not about the model. They are about the environment — network mocking, tool state, and context bleed between runs. Here is how to fix it.