← Back to Writing
·3 min read

Delegation UI

uiagentsagentic experience

"Frontend/UI is dead" is a sentiment that's been steadily growing with each new AI coding tool that ships (Hell for any profession these days). For sure, for a certain kind of frontend work, the writing is on the wall (think translating Figma files into code).

But I think that framing misses what's actually happening and needs to be properly articulated.

What's actually dying is navigation-first UI; the assumption that an interface exists to help a user drive software manually. That assumption held for thirty years but it no longer holds with Agents.

Looking back, there have been two dominant UI paradigms that have defined how humans interact with software. We're now watching a third one emerge in real time, this is their timeline:

Command line UI (1960s-1984): You told the machine exactly what to do, step by step. The interface was literally just a prompt.

Navigation UI (1984–present): You drove software through menus, flows, and screens. The interface was a vehicle. This is the paradigm that held for thirty years and the one that's actually dying.

Delegation UI (2023–): You describe an outcome and an agent pursues it. The interface is no longer a vehicle you drive; it's a cockpit you monitor. Clicking through steps is going away, instead you're setting intent, reviewing decisions, and maintaining oversight.

Unlike the previous two, nobody has thirty years of established patterns to borrow from. We know how to build navigation UI in our sleep.

You know what a button does, you know where the menu lives, you know what a confirmation dialog is for. Every component library on earth was built for this world.

Delegation UI has none of that. There's no equivalent vocabulary, no shared answer to what "a thing that lets you trust an agent" even looks like.

That gap is the problem. Now that agents move faster than we can follow, the interface has to carry weight it's never had to carry before: communicating confidence levels, surfacing decisions that need human review, letting users intervene without breaking the flow. Navigation UI was never asked to do any of this.

This is the problem I've been building toward with depute: an open source React component library for the delegation UI.

The component list is shaped by how much autonomy an agent has at any given moment: the more autonomous, the more the interface needs to compensate with visibility and control.

Primitives like:

These primitives aren't new UI patterns for novelty's sake. They're the first components that delegation UI actually needs; for interfaces where the user supervises instead of operates.

The field is early and the components don't exist yet. So I started building them.

depute.dev open source, zero dependencies.