Hybrid Human–Agent Software
One product, two complementary operators: people work directly where it helps, agents handle the rest.
Emlyn O'Regan
A hybrid application is designed for both a human and an agent to operate. It isn't merely a conventional interface with a chatbot attached. Both operators can act on compatible underlying state and can alternate without a costly handoff.
Different tasks suit different operators. A human working with audio may prefer to see a waveform, drag a selection and listen to the result. An agent may be better at batch-processing clips, adjusting detailed parameters and repeating an operation across many files. A visual website editor offers a similar division between direct layout work and structured changes made by an agent.
The essential property is complementarity, not identical access. An agent does not need to reproduce the human interface, and a human does not need a visible control for every machine operation.
A path to simpler products
Early hybrids might preserve a comprehensive UI and add agent tooling. Mature hybrids could remove infrequent operations from the human interface, leaving a smaller set of excellent direct-manipulation tools.
The human interface might itself become partly generated on demand, with controls assembled for the task at hand.
What makes this difficult?
Both operators need a dependable understanding of what changed. The agent must not silently overturn deliberate human work; the UI must reflect agent changes. Undo, provenance, permission boundaries and predictable state matter more than the choice of protocol.
Hybrid is therefore a design commitment: human and agent are peers using different interaction methods, not a human operating an app with an occasional automation script.
Related: Three Modes of Application Interaction · The different 10% problem.