The Different 10% Problem
Each person may use a small fraction of a product's features, but not the same fraction as everyone else.
Emlyn O'Regan
A common observation about complex software is that most people use only a minority of its functions. The exact percentage isn't the point; the crucial detail is that different users rely on different minorities.
An audio editor's noise-reduction tools may be essential to one person and irrelevant to another. A reporting application's unusual export format might matter enormously to a small group. Removing rarely used features can therefore reduce complexity for the majority while making the product unusable for others.
This is why conventional interface simplification runs into a ceiling. The vendor has to keep capabilities accessible across a heterogeneous audience.
The hybrid alternative
An agent can expose a capability when a person needs it, without requiring every user to navigate its controls all the time. Frequent direct-manipulation tasks remain visible. Rare tasks can be invoked through conversation or temporary interfaces.
Eventually an agent could do something stronger: supply a person's required capabilities from separate products, rather than accepting the vendor's predetermined bundle.
That turns a problem of interface design into one of software architecture and economics.
A human-facing workspace might show the user's own frequently used controls, perhaps with small panels generated on demand. The remaining operations are accessible through the agent.
There's a caveat: people sometimes discover valuable features by noticing them in the interface. Agent-mediated discovery must be helpful without making the user dependent on opaque recommendations.
Related: The Disappearing Interface · Software Slivers · Agents as Software Discoverers.