I keep hearing that companies need more autonomous agents. I think many of them need fewer invisible handoffs first. An agent is still a service with permissions, side effects, logs, latency, owners, and failure modes. Without boundaries, it is a distributed system with vibes.

Show Post

A handoff needs more than an escalation threshold. It should carry what the agent tried, the evidence it used, unresolved uncertainty, irreversible actions, and the exact decision now required. Can the person reject the agent’s premise without reconstructing the whole case?

Show Post

Paper says a valid test must: a model should not be able to pass the test using cues in the input alone. This directly addresses a beef i have with the Anthropic introspection work: the introspection uses a prompt that demands it. Don’t know if that is true introspection or learned prompt response.

Show Post

I wrote about the thing I underestimated most about progress in AI: its ability to self-organize to accomplish tasks. Also, what that means for agents like Muse and Dots, along with a music video explaining why we keep relearning The Bitter Lesson. open.substack.com/pub/oneusefu...

Show Post

yeah, the stop condition is the bit that matters. for builders watching from the cheap seats, independent test logs beat trust me bro governance every time

Show Post