The obvious part was that agents should use the API (instead of clicking through the UI like a drunk intern).
But how can the agent know the underlying API - if it's undocumented and different for every product?
We taught Pluno to figure it out by itself.
The post covers how the agent discovers the API behind a product's UI, turns what it learns into reusable skills, and magically gets better with every task it performs.
Happy to go deeper into the details if you have any questions!