I'm running an experiment with a team of Claude agents to build a decision support app for biotech investors. Regardless of the results, the agents simply do too much for me to monitor what's going on. Standard updates to Slack are ok but that format lends itself to short-and-sweet posts.

Podcasts turn out to be a pretty good way to keep track of progress. At the end of a unit of work, the agent writes a script from the agent chat history (including my interjections) and generates the audio with voices from Eleven Labs.

A funny example came from an episode where I learned about the team's plan to run a large test suite. Instead of going through my Claude plan the team thought it was a good idea to use API calls. Estimated cost: $730.

I didn't get banned but you can imagine my mature, productive response when I hopped into the chat to stop the test.