We build the systems that make agent autonomy trustworthy.
Unsupervised started in 2017 automating analytics for enterprises like AT&T and Disney, where an agent has to get the numbers right. That work showed us where autonomy breaks. Now we publish the harnesses, workflows, and checks we built to keep agents working without a person watching every step.
What we believe
01
Autonomy has to be earned.
An agent gets more room to work when the system around it makes boundaries, checks, and evidence explicit. Trust is something you build in the harness, not something you assume.
02
Humans handle judgment.
People set direction, approve risk, and decide the edge cases. They should not have to watch every step of a run to know it went right.
03
Agents show their work.
A run should expose what it changed, why it believes the result, where it is uncertain, and what still needs review.
Investors
Hiring
Work on these problems with us.
If you want to work on agent reliability and control, email requests@unsupervised.com. The strongest introduction is a link to open-source software you've built around making agents reliable or controllable.