Design Just Got Promoted

The claim that design is lost contains a category error. Drawing got cheap. Deciding got more valuable, and there is more to decide than at any point in the last twenty years.

Falk Gottlob7 min readNew

The short version

The claim that AI killed design confuses design with drawing. Drawing got cheap. Deciding did not, and there is more to decide now than at any point in the last twenty years, because every AI feature shipped since 2024 created interface problems that did not exist before. The shift is from producing artifacts to specifying behavior: constraints with failure conditions, a defined range instead of a final state, judgment written into a rubric that scores work without you in the room, and the wrong path owned as a first-class surface. Six operating rules follow, and there is a worksheet at the end you can hand your team on Monday.

The most valuable decade in design's history started about eighteen months ago, and a large part of the field is spending it writing eulogies.

You know the post. Design is dead, the craft is gone, the juniors have no path, screenshot of a layoff announcement, nine thousand reactions. It performs because it is easy to write and impossible to be wrong about. Nobody ever gets fact-checked on a mood.

But the claim has a category error sitting right in the middle of it, and once you see it you cannot unsee it. It confuses design with drawing.

Drawing got cheap. Deciding did not. And there is more to decide right now than at any point in my career.

What actually changed

I have run product and design, research included, for more than ten years, across companies where the interface was a marketing site and companies where the interface was a clinician making a call about a patient. In every one of them the same bottleneck showed up: we could not produce options fast enough to learn what was right.

That bottleneck is gone. Permanently. A team can now put twelve credible directions on the wall by Wednesday. I wrote about what that does to a build cycle in Instant Prototyping, and the same collapse is now happening one layer up, to design itself.

Look at what that does to the org. It does not remove the need for a designer. It removes every excuse the company had for not knowing which direction is correct. When options were expensive, "we went with the one we had time to build" was a reasonable answer. Now it is a confession. Somebody has to say why this one and not those eleven, and say it in language a team and an agent can both execute.

That somebody is a designer. That job used to be maybe fifteen percent of the week, hidden inside the making. It is the whole week now.

That is not a demotion. In any other industry we would call it moving from the factory floor to the spec.

The surface count went up, not down

And the doom take misses the biggest change of all.

Every AI feature shipped in the last two years created interface problems that did not exist before. What does the product do when it is unsure. How does a person see what the system already did on their behalf. What is the shape of an undo when the action was taken by an agent at 3am. How much of the reasoning do you show, and to whom, and when does showing it destroy trust rather than build it.

None of that was in a design system in 2023. None of it is solved. Most of it is not even named yet, which means whoever names it well gets to define it for everyone else.

There has never been a better time to be the person who is good at this. The open problems are enormous and the field is standing around arguing about whether it still exists.

The shift, stated plainly

From producing artifacts to specifying behavior.

One sentence, and everything below follows from it.

An artifact shows one state and requires you in the room to explain the rest. A behavior specification covers every state, including the ones you did not imagine, and it works while you sleep. One scales to the number of screens you can personally look at. The other scales to the number of screens your company can generate, which is now a much larger number.

Designers who make that shift are about to be the most leveraged people in their companies. Not eventually. This year.

The operating rules

Handbook section. Give these to your team on Monday.

1. Write the constraint before anything gets made. Not a principle. Principles are decoration, and "delightful, human, trustworthy" has never once stopped anyone from shipping anything. A constraint is a rule with a failure condition attached, so you can tell when it was broken. Any suggestion a user can accept in one click needs an undo that is equally cheap. A confirmation cannot be the first time someone learns what the system already did. Those are checkable by a person, by a reviewer, and by an eval.

2. Design the range, not the screen. When output is assembled at runtime there is no final state to draw. Define the best case, the worst acceptable case, and what the product refuses to do. Everything in between is allowed and does not need your review.

3. Put the judgment in a rubric so it executes without you. Your taste currently lives in a Thursday meeting. Move it into something written, scored, versioned, and argued about. This is how craft scales past one person's calendar, and it is the single highest-return thing a design leader can do this quarter. The mechanics are the same ones in The Eval Is The Spec, pointed at interface quality instead of model output.

4. Own the wrong path as a first-class surface. Recovery, disclosure, confidence, correction, trust repair. In a deterministic product the unhappy path was rare, so we gave it to whoever had capacity. It is common now, and it is where the whole relationship with the user is either built or lost. Claim it before someone else does.

5. Spend the freed hours on evidence, not on more options. The generation constraint is gone. The knowing constraint is the only one left. Teams that use their new speed to produce more untested directions have converted a real advantage into noise.

6. Move design upstream of generation. If your review is happening after things are made, you are inspecting output like a QA function. Set the bounds first and the review becomes a ten-minute conversation about exceptions instead of a two-hour tour of work that is already done. Same move PM as Editor describes on the product side, arriving now for design.

For the leaders reading this

Your team is anxious because the field's loudest voices are telling them a story with no role for them in it. You cannot fix that with reassurance. Reassurance is just the optimistic version of the same passive frame.

Fix it with a job description. Hand them the six rules above, pick the two that matter most for what you are building, and give someone ownership of each. Anxiety about the future dissolves fastest when there is a specific piece of it you have been made responsible for.

Then do one thing yourself this week. Take the last design decision you had to argue for, and write the three constraints that would get a competent stranger, or an agent, to that same decision without you in the room. Send it to your team. That document is the first page of the spec your practice is going to run on for the next ten years, and almost nobody has started writing it.

Good time to be early.

The six rules with the failure condition, the check, and a blank table for each, plus a two-week adoption plan, are in The Design Operating Rules. Built to be usable by someone who never read this chapter. If you need the short version to send to a skeptical exec, that one is What do designers do when AI generates the screens?


First in a series on the design operating model. Next: why you cannot mock a distribution, and what replaces the screen as the unit of design.

Share this post

Take the template

The six operating rules

Download

Next up

One chapter a week, 26 in total. See the full arc and what is coming next.

Frequently asked

Is AI making product design obsolete?+

No. It made drawing cheap, and drawing was never the whole job. A team can now put twelve credible directions on the wall by Wednesday, which means somebody has to say why this one and not those eleven, in language a team and an agent can both execute. That work used to be about fifteen percent of a designer's week, hidden inside the making. It is the whole week now.

What does a designer actually do when the screens generate themselves?+

Specify behavior instead of producing artifacts. Write the constraints before anything gets made, define the range instead of the final state, move taste into a rubric that scores work without you in the room, and own the wrong path: recovery, disclosure, confidence, correction. None of that was in a design system in 2023 and most of it is still unsolved.

What is the difference between a design principle and a constraint?+

A principle has no failure condition. Delightful, human, trustworthy has never once stopped anyone from shipping anything. A constraint is a rule you can catch being broken: any suggestion a user can accept in one click needs an undo that is equally cheap, or a confirmation cannot be the first time someone learns what the system already did. Both of those are checkable by a person, a reviewer, and an eval.

Why can't you just draw the screen for an AI feature?+

Because there is no final state to draw. Output is assembled at runtime, so the artifact shows one sample of a distribution and requires you in the room to explain the rest. What you specify instead is the best case, the worst case you would still ship, and what the product refuses to do. Everything in between is allowed and does not need your review.

How should a design leader handle a team that is anxious about AI?+

Not with reassurance. Reassurance is the optimistic version of the same passive frame. Hand them a job description instead. Pick the two operating rules that matter most for what you are building, give each one a named owner and a deliverable inside two weeks. Anxiety about the future dissolves fastest when there is a specific piece of it somebody has been made responsible for.

Which design problems are still unsolved in AI products?+

How a product says it is unsure without looking broken. How a person sees what the system already did on their behalf. What undo means when an agent acted at 3am. How much reasoning to show, to whom, and when showing it costs trust instead of building it. Most of these are not even named yet, which means whoever names one well gets to define it for everyone else.

Related reading

Chapters and essays on the same thread, across both handbooks.