Figma's Trial Closed 70% of the PM Gap. The Gate Is the Rubric, Not the Designer.

Jakob Nielsen's roundup this morning carries the first randomized trial of a prompt-to-design tool I have seen, and his advice to managers at the end of it is the sentence I want to argue with.

The trial is Figma's own, run on Figma Make, with an independent firm moderating. The result is clean. Product managers with the tool got 35% faster at design tasks. Designers got 17% faster. The gap between the two roles mostly closed.

Nielsen's advice: expect more design work to originate outside the design team, and make designers its quality gate.

I have run product next to design at Adobe and Salesforce. A person as a gate is the thing I would not do, and the trial is the reason why.

The short version

Figma's randomized trial of Figma Make, as Jakob Nielsen reports it on October 19, 2026, put 50 product managers and 50 product designers through the same three design tasks and found AI cut PMs' time 35% and designers' 17%, closing roughly 70% of the gap between the roles. Nielsen tells managers to make designers the quality gate for the design work PMs will now produce. A person as a gate caps the product at what one person can look at, and the trial just raised the volume. The ownership map in my design handbook gives design six of thirteen rows, none of them "review the PM's prototype." The gate is the rubric. Design writes it, a PM's prototype arrives scored, and the designer's time goes to the rows no rubric can hold yet. The sizes are the ones I already use instead of a PRD: a prototype in four to six hours, a one-page rubric with three to five criteria and at least 20 test cases, graded binary per dimension.

What the trial measured

The paper is by Remy Stewart, Olabode Anise, Andrew Hogan, and Augustus Griffin at Figma. MeasuringU recruited 50 product designers and 50 product managers and moderated the sessions. Half of each group got Figma Make after a five-minute primer. The other half designed by hand in regular Figma.

Everyone did the same three tasks on a social media app: convert a screen to dark mode, add a Help & Support item to a settings menu, and prototype a comment flyout. Each task came with a picture of the finished design, and the clock stopped when a moderator confirmed a match.

Averaged across tasks, AI cut completion time about 20%. The gains were lopsided. PMs saved 35%. Designers saved 17%, and only on the flyout, the hardest task. AI made PMs 38 percentage points more likely to finish the flyout at all. By Nielsen's arithmetic from the paper's regression, PMs working by hand needed 26 minutes more than designers to finish all three tasks; with AI on both sides, 8 minutes more. Roughly 70% of the skill gap gone.

Nielsen names why designers gained nothing on the easy edits: delegatency, the fixed cost of writing the prompt, waiting, and checking. An expert adds a menu item faster than she can describe it. Only on the flyout, which took designers 24 minutes by hand, did the tool's speed outrun its overhead.

He is also careful about what the study is. It measured how fast people reproduce a known design, "where AI is strongest and designers' edge is thinnest." In real work nobody hands you the reference image. His line: deciding what the design should be is the design job.

All of that I would hand to a design lead unchanged.

The clause I would replace

Then the prescriptions. PMs should show ideas as prototypes instead of describing them in specs, and let a designer judge them. Designers should stop selling tool speed and sell judgment. And managers should "expect more design work to originate outside the design team, and make designers its quality gate."

The first two are already standing rules on this site. Kill the PRD. The Prototype Is the Spec. made the first one the rule in May. The second is the thesis of the design handbook.

The third clause is the problem, and it is a problem of arithmetic rather than of taste.

A gate is a person who looks at everything before it ships. The number of things that person can look at is fixed by their calendar. Before the trial, the design lead was already the only quality gate in most companies I have worked in, and the product was already capped at the number of things she could look at in a week.

Now read the trial's numbers from the other side of the gate. The people who used to be 26 minutes slower than designers are 8 minutes slower. They are 38 points likelier to finish the hard task at all. The volume of design work arriving from outside the design team is about to go up, which is exactly Nielsen's first sentence to managers. His second sentence routes all of it through the one resource that did not get faster.

What Design Owns Now(coming Nov 11) has the line I keep coming back to: "The failure mode without one is a design leader who is the only quality gate in the company, which caps the product at the number of things one person can look at." I wrote that before the trial existed. The trial is the number that makes it urgent.

Six rows, and none of them is review

The ownership map in that chapter has thirteen rows. Every row has exactly one owner, a function and then a name under it. Design owns six:

1. What the product will not do.
2. The constraint set for a surface.
3. The acceptable range of generated output.
4. The quality rubric and its dimensions.
5. When output is good enough to ship.
6. The wrong path: recovery, disclosure, correction.

PM owns the problem, the users, the instrumentation, and the sequence. Engineering owns the mechanism and the budget.

None of the six design rows says "review the PM's prototype." Row five looks like it does, and the difference between row five and a gate is the whole argument. Design owns the call on when output is good enough, and the rubric in row four is what makes that call reviewable instead of personal. If the rubric exists, the call runs without the designer in the room. If it does not, the designer is the call, and the queue forms.

The Rubric Is the Spec says it the other way round: taste that lives in a recurring meeting caps quality at what one person can personally review. A rubric is that taste in a form that runs without her.

The gate is the rubric, and design writes it

So the rewrite for Nielsen's third sentence: do not make designers the quality gate. Make the rubric the gate and hold design accountable for writing it.

What changes on a Tuesday. A PM builds a prototype in Figma Make. Before it reaches anyone, it gets scored against the rubric for that surface. The score travels with it. The designer sees a prototype with a 4 of 5 and one failed dimension, not a Slack message asking for thoughts. Her time goes to the failed dimension, and to the rows no rubric can hold yet, which Nielsen correctly says is the real design job: deciding what the design should be when nobody handed you the picture.

The sizes I use, from the rule that replaced the PRD here: a working prototype built in four to six hours, real enough that a customer can use it for five minutes. A one-page rubric with three to five criteria and at least 20 test cases. Done is when the rubric passes the threshold, not when a document is approved.

Two refinements from How I Build Eval Rubrics, because a rubric written from first principles is a rubric nobody opens again. Grade thirty outputs before you write a single criterion, so the dimensions come from what actually went wrong. And go binary per dimension wherever you can. A 1-to-5 scale hides disagreement inside the 3. Pass or fail per row forces the argument into the open, where it belongs.

Nielsen's own numbers set the pace. The flyout took designers 24 minutes by hand. A rubric for a surface takes longer than that to write well and seconds to apply. One afternoon of design time buys a gate that does not queue.

What the trial could not test, and why that is the point

Nielsen closes with the caveat that makes the argument for me. Every task in the trial came with a picture of the finished design. Someone had already decided which pallet to lift and where to put it, in his forklift metaphor. Take away the picture, he predicts, and the gap between PMs and designers reopens exactly where it matters most.

He is right, and the picture has a name. It is the rubric, written before the prototype exists. The trial did not have to measure judgment because it supplied the judgment in the box. In a real team, the rubric is what goes in the box, and writing it is the design work that does not get faster with a prompt.

That is also why this belongs to the Product Leadership argument rather than to a tools debate. The product org chart was calibrated to a build cost that collapsed. The trial measured one piece of that collapse to the minute. The leadership work is deciding which coordination role stops being necessary, and the design lead as the review queue is one of them. Part 1 of the Design Transition made the same case from Figma's own ratio, 17x the code and 1.3x the releases: generation stopped being the bottleneck, and the spec became the deciding end. This post is the org rule that follows from it.

This week: pick one surface your PMs will prototype this quarter. Ask design for its rubric by Friday, three to five rows, binary where possible, twenty test cases. Then count how many prototypes arrived scored. If the answer is zero, you still have a gate, and it has a calendar.

Related answer: Should designers be the quality gate for PM prototypes?

Sources: Jakob Nielsen, "UX Roundup: AI by Design Phase | Product Managers vs. Designers | Individualized Tutoring | Text Labels | Thumbnails | Stopping AI | Unsubscribe | OODA Loop | Timeouts | Buzzwords | Registration," UX Tigers, October 19, 2026, the "AI Design Tools Help Product Managers Twice as Much as Designers" section read in full; the trial's figures are cited as he reports them. Remy Stewart, Olabode Anise, Andrew Hogan, and Augustus Griffin, "Does AI Save Time on Product Design? A Randomized Controlled Experiment of AI Prompt-to-Design Workflows," arXiv 2609.26725, September 22, 2026, abstract read. Falk Gottlob, "What Design Owns Now(coming Nov 11)," "The Rubric Is the Spec," "Kill the PRD. The Prototype Is the Spec.," and "How I Build Eval Rubrics."