Evaluating Agents
For agents, the trajectory is the test
Self-improving agents are an attractive idea
The governance implications are uncomfortable
Understanding is the new moat
Do you understand what you produced, and can you defend it when it's wrong?
Seoul is a vibe today
Accountability is more than I think it is
Proving what an agent did: my Future of Privacy Forum panel
The open question is who holds the record once the agent acts.
ICML: The call is from inside the model
The scientists and academics here are starting to work on judgment.
The evidence layer is empty: notes from the NYU fiduciary AI workshop
The duty stays with a human or a company, not the AI.
Your AI reviewed the contract. But you can't cross-examine a log
As agents start doing real legal work, we have no clean way to prove what they actually did
