What actually improves an analytics agent in production?
I shared some of the experiments, missteps, and user feedback that shaped how we built Hive Mind, an internal agent at Robinhood. Would love to compare notes with others working on similar systems: https://t.co/MCgBrdMLOs
Playing around with TypeSafe AI’s Jev. I created a decision layer for data agents to act on. It ended up being ~4x faster and ~2x cheaper for automating BI analysis.
In a policy-bound 120-case run:
- Jev: 953 ms, 100% recall.
- Luna: 3,383 ms, 90.3% recall, 14 unsafe actions.
Slop-Creep made it to TLDR PM Newsletter!
Granted, the newsletter focuses on NOT reading underlying articles. Delivered to over 400k inboxes with only 400 people opening my article. At least the concept of slop-creep is getting shared 😅
https://t.co/gyBD7D7GJc