Dianne Penn joined Anthropic in 2023 as its first technical product manager, when the entire product team was five engineers. She is now Head of Product for Anthropic's AI Research and Labs teams. In that time she helped ship every model from Claude 2 through Fable and incubated Claude Code, MCP, computer use, tool use, and reasoning. Before Anthropic she built AI at Amazon Alexa and traded high-yield bonds at JP Morgan Chase.
The interview covers the specific mechanics behind how Claude got good at coding, the eval-driven development loop Penn's team runs, and why Claude's willingness to push back on users is treated as a feature, not a bug. These are not high-level strategy takes. Penn explains the actual process: how evals are constructed, what signals matter, and where human judgment still cannot be automated away.
If you work on AI product, the section on 'token maxing' and the jagged edge of model capability is worth your time. Penn is describing patterns that most teams discover the hard way, months after deployment. The full conversation is on YouTube, Spotify, and Apple Podcasts.
[READ ORIGINAL →]