Building Products for Humans and AI Agents

For decades, software had one kind of user: a person clicking, typing, and reading. That assumption is breaking. Increasingly, the "user" acting inside a product is an AI agent, and it interacts nothing like a human. Building for both at once is one of the most interesting design challenges we face at Besenes, and we think it's one every product team will face soon.

Two users, very different needs

Humans and agents want almost opposite things from an interface.

People need visual hierarchy, context, and reassurance. They scan, they hesitate, they benefit from good defaults and gentle guidance. They read a screen.

Agents need structure, clarity, and predictability. They don't scan a page; they parse it. They want well-defined actions, explicit state, machine-readable data, and stable contracts that don't shift underneath them. Ambiguity that a person navigates intuitively is exactly what trips an agent up.

Design for only one and you shortchange the other. A beautiful interface an agent can't operate is a dead end; a machine-friendly API with no human affordances leaves your people behind.

Principles we design by

Building for both audiences forces some discipline. A few principles guide us.

• Structure first. If the underlying model of the work is clean, clear entities, actions, and state, both a human UI and an agent interface can sit on top of it. Structure serves everyone.
• Actions, not just screens. Every meaningful thing a person can do should be an explicit, well-defined action an agent can take too, with the same rules and permissions.
• Shared context. Humans and agents should work from the same source of truth. When they diverge, coordination breaks down.
• Legible state. What's happening, who did what, what's pending, should be visible to a person and queryable by an agent.
• Guardrails that apply to both. Permissions and boundaries shouldn't care whether the actor is a person or an agent. The rules are the rules.

Collaboration, not just coexistence

The point isn't to bolt an agent mode onto a human product. It's to design a space where people and AI agents genuinely collaborate: a person sets direction and approves, an agent executes, and the handoffs between them are smooth because both are first-class from the start.

That is exactly what we're building toward with Besenes. The Work Hub is designed as shared ground for people and AI workers, delegation and approvals make the handoffs explicit, and company memory gives both sides the same context to act on.

The products that thrive in an AI-native world will be the ones built for both audiences deliberately, not the ones that treated agents as an afterthought. See how Besenes does it, or read more on the blog.