Two AI agent platforms, run side by side for a fortnight. Different bets, both legitimate, neither obvious.
Both treat agents as something more structured than a chat. The differences are in what they assume the agent should be doing with its time.
Dust leans toward agents-as-coworkers — a small team of specialists, each one pointed at a particular data source or task, configurable without writing code. It is the platform you would choose if you wanted multiple non-engineers to ship agents that read from the same shared brain.
Hyper Agent leans toward agents-as-operators — the agent has hands. It can drive a browser, click through screens, fill in forms, deal with software that does not expose an API. That is a different problem than retrieval, and a different platform shape.
The studio uses both, for different jobs. The mistake is asking either one to do the other's job.
Filed Apr 2026 · Folkestone