The thing most AI teams aren't tracking yet
Been speaking to a lot of teams running AI agents in production over the past few months. Sales agents, coding agents, research agents, customer support bots all live, all handling real work.
One question I started asking: do you know what your agent did yesterday?
Not the output. The actions. Which tools it called. What data it sent where. What decisions it made without a human seeing them.
The answer is almost always no.
And it's not because the teams are careless, they are smart people. It's because the tooling for agent observability is miles behind the tooling for agent capability. We got really good at building agents before we got good at understanding what they do.
I think this is going to become a serious problem for organizations at scale, not a compliance or legal problem, an operational one. When an agent makes a bad call at 3am and you can't trace why, that's a business problem.
Curious if others are seeing this pattern. How are you handling visibility into what your agents are actually doing?
(We are building in this space — happy to share what we have learned if it's useful)