Conference hallways stopped arguing about whether AI works. The new fights are about agents that act safely, evaluations that mean something, and products people can trust.

Agents stole the keynote — evals stole the hallway
Demos of autonomous agents looked impressive on stage. Offline, engineers asked harder questions: How do you test an agent that chooses its own tools? What does regression mean when the plan changes every run?
Those questions matter more than another flashy demo. Production AI lives or dies on evaluation quality.
Trust is now a product feature
Talks on transparency, citations, and human oversight drew packed rooms. Buyers and builders alike are tired of black-box answers dressed up as intelligence.
The cultural shift is clear: show your work, measure your failures, and design for escalation.
Bring the debate home
After the event, the useful ritual is a short internal write-up: what we will try, what we will ignore, and what we will measure next quarter. Conferences create ideas. Teams create outcomes.
Topics
- AI conferences
- AI agents
- LLM evaluation
- trustworthy AI
