For thirty years, software has run on one contract: you click, it responds. We designed the click — the button, the field, the flow, the exact spot where human intent met deterministic logic. The whole discipline grew up inside that contract.
Agents tear it up. An agent doesn't wait for the click; it acts on the intent behind it. You stop operating software and start delegating to it. And the instant a system can do real things on your behalf — book, buy, write, decide, call other systems — the question stops being "how does this screen work" and becomes "how much do I trust this, and how do I stay in control of what it does in my name."
That isn't a new component library. Three things change.
Intent replaces controls
Direct manipulation rewarded precision: the right control doing exactly one thing. Agents reward expression — you describe an outcome and the system finds the steps. So the new craft is helping people say what they want well: scope it, bound it, correct it, without dumping them back into the manual controls the agent was supposed to retire. The best agentic interface feels less like a cockpit and more like a sharp brief to a capable colleague.
Legibility beats polish
Deterministic systems earn trust through familiarity. Probabilistic ones earn it through legibility — I can see what the agent understood, what it's about to do, what it did, and how to undo it. A gorgeous screen that hides the model's reasoning is worse than a plain one that shows it. I spend more time now on the seams — preview, confirm, provenance, the "here's what I'm about to do" moment — than on any hero shot. In the agentic era, undo and the audit trail are core UX, not edge cases.
Design the loop, not the screen
Screens are stops; agents run in loops — plan, act, observe, correct. Designing one screen at a time misses where the experience actually lives: the hand-offs between person and system over time. Who has the initiative right now? When does the agent ask versus assume? How does it fail — because it will? Designing the loop means designing for recovery, not just the happy path. The failure modes are the product.
The Great Recomposition
This shift doesn't just change the interface — it changes the people who make it. I co-authored a piece called The Great Recomposition about exactly this. For years the worry was decomposition: AI carving the design role into ever-thinner slivers until there was nothing left to do. What's actually happening is the opposite. The slivers are recombining. When a designer can describe an outcome and have the system build it, the old walls between design, engineering, and product stop paying rent.
The designers who thrive in the agentic era look less like specialists and more like builders, intent architects, and generalists. Builders, because the distance from idea to working artifact has collapsed — you can ship the thing instead of speccing it for someone else. Intent architects, because the scarce skill is no longer pushing pixels but framing the problem precisely enough that a capable system can run with it — naming the outcome, the constraints, the guardrails, the definition of done. And generalists, because value now lives in the seams between disciplines, and the people who can move across the canvas, the codebase, and the company are the ones who can actually compose a whole experience rather than hand off a fragment of one. That arc — IC craftsperson to architect of intent — is the one I've been living, and the one I'd bet on.
Why better models won't save us
It's tempting to wait for this to be a model problem that scale solves. It won't be. A stronger model raises the ceiling on what an agent can do and the stakes of getting it wrong. Capability without legibility and control just makes confident software people are afraid to use. The teams that win the agentic era won't have marginally better benchmarks — they'll have made powerful systems feel trustworthy and steerable. That's a design problem, and it's the best one our field has had in a generation.
I get to work it from both sides — leading design for agentic data products at Google, and building AI ventures where the whole company is the experiment. The throughline is old-fashioned: technology earns its place when it makes people more capable, not less in control. Agents raise the ceiling on capability. Our job is to make sure control comes with it.
A working note — I revise it as the work teaches me more. Reactions welcome: i.wooten@gmail.com.