The Morse Blog
Using a model to grade your agent's output is useful and cheap to start. Here is how to keep it honest: sample, calibrate against people, keep it apart from hard outcomes, and watch what it costs.
Read post →A single blended number for agent cost and quality feels tidy, but it hides the tradeoffs you need to see. Here is how to read them next to each other instead.
Read post →Agents moved from demos to production faster than the tools for watching them. What the industry posts will cover, and the few positions we start from.
Read post →