Everyone talks about AI agents. Hardly anyone really shows them.
Last week I set up two AI agents on two separate computers. Gerda runs on my Windows server. Kelly on a Mac mini.
I gave them an assignment: "A mail does not reach a customer, but our ERP says 'sent'. Find out why and fix it."
And then I let them talk to each other.
Gerda dug into the data. Kelly challenged her findings. They went back and forth, every step live in a group chat where I simply watched. No human stepping in.



Result: they found the root cause (a wrong e-mail address the system never flagged as an error) and fixed it. By themselves.
This runs on Claude Opus 4.8 with the new Workflows: agents split up work, check each other, and converge on a solution together.


This is not a thing of the future. This ran today.
The full demo is on its way. Follow Printelligence.