# AI design adoption: measure review delays and rework, too

More AI-generated designs do not necessarily mean faster delivery. Find out whether adoption reduces the review delays and rework that slow your team.

By Sentia Labs

Published: 2026-08-21
Updated: 2026-09-17

Canonical: https://www.sentialabs.ai/blog/output-is-up-outcomes-are-not

[Figma's 2026 AI report](https://www.figma.com/blog/2026-ai-report/) describes AI changing collaboration and expanding participation in design and development. It raises a practical question for leaders: is that activity improving customer and business outcomes?

The report's public summary does not establish that products are getting better or worse. The distinction matters. Adoption is an observed behavior; product quality and business impact need their own evidence.

## What the published findings support

Figma describes a research program spanning three years, with 8,403 total survey responses and 639 qualitative interviews. Its 2026 coverage spans ten markets.

The published findings show more designers participating in development, more developers doing design, and 41 percent of respondents saying AI has meaningfully changed how their teams work together, compared with 7 percent two years earlier. Ninety percent say design is at least as important as before AI.

Figma also describes friction when individual adoption and organizational direction are disconnected. Our interpretation is that design leaders need to examine the operating model around AI, as well as the tools themselves.

## Abundance does not create alignment, it prices it

Here is the mechanism, and it is not specific to design.

Every organization has coordination overhead: the cost of getting a group of people to agree on what to build and then build the same thing. When production is slow, that overhead is hidden, because coordination happens in the gaps while people wait for work to finish. The seams are there. They just are not on the critical path.

Consider a workflow where producing a first draft becomes much faster. Now coordination is the critical path, and every seam that used to be absorbed by slack becomes a visible delay. The organization did not get worse at aligning. Alignment simply became the bottleneck, and it was never designed to be one.

This is one way additional output can fail to produce better outcomes: the constraint moves, while the operating model stays the same. Whether it is happening in a particular team needs to be measured.

## What this means specifically for design organizations

Design gets a concentrated dose of this for a structural reason: it sits at the seam. Design work is where product intent, customer evidence and engineering feasibility have to reconcile. When the boundaries dissolve and more people participate in design, the reconciling work grows, and it grows fastest for the person accountable for coherence.

That person is the design leader, and their instrumentation did not change at all.

They are now responsible for more surface area, produced by more people, in more places, at higher velocity, with the same tools they had when the org was half the size and the boundaries held. The gap between what they are accountable for and what they can observe has widened every quarter for two years.

## The trap: measuring adoption

The natural response is to measure AI adoption. Seats, active users, prompts per week, percentage of the team using the tool. Every one of these is easy to collect, which is why every one of them gets collected.

None of them answer the question. Adoption tells you a tool is being used. It says nothing about whether the work got better, so usage and improvement should be measured separately.

Worse, adoption metrics create pressure in the wrong direction. A team told to increase AI usage will increase AI usage, including in the places where it adds rework. One failure mode to watch for is when AI-accelerated first drafts push more decisions downstream, so a team ships faster into review and then spends the saved time in revision cycles. Net velocity is flat and everyone feels busier. Adoption metrics show that as a success.

## What to measure instead

Three things, and none of them mention AI.

**Cycle time by stage, not in aggregate.** Total time from brief to shipped is a summary statistic that hides the interesting part. The useful view is where the time sits: ideation, critique, review, handoff, rework. When AI compresses one stage and inflates another, only the staged view shows it. The aggregate can stay flat while the shape of the work changes completely.

**Rework rate.** How often does something come back after it was considered done, and at which stage did it come back from? This is the single most direct measure of whether faster production is producing better decisions or just earlier ones. It is also the number that catches AI-accelerated slop, because slop shows up as rework, not as slower output.

**Decision-to-outcome linkage.** Which decisions shipped, which priority did each serve, and what changed. This is the measure that survives contact with a business review, and it is the one that says whether any of the additional output reached the business.

Notice what these have in common. They are all measures of the organization's operating quality, not of tool usage. The question is never "are we using AI." It is "did the way we work get better, and can we tell."

## Review a tool rollout as an operating change

Choose one comparable workflow and record a baseline before the rollout. Follow the work through brief, exploration, critique, validation, handoff, and QA. Measure waiting time and rework as well as production time.

Review active usage alongside license cost and onboarding effort. An idle seat may call for training, reassignment, or removal; those are different actions and need an owner. A popular tool may still create more downstream repair than it removes.

Set a follow-up date and ask whether the workflow improved enough to justify expanding access. Treat Figma's findings as industry context, not proof of a result in your organization.

## The uncomfortable conclusion

If output is up and outcomes are flat, the additional output has negative value. It consumed real time and attention, and it produced coordination load rather than results.

That is not an argument against AI in design work. Broader participation can make design judgment and shared standards more important. It is an argument that the constraint has moved from making things to deciding what to make and getting a larger, blurrier group of people aligned behind it.

Organizations that keep optimizing production will keep widening the gap. The ones that instrument how they actually operate, and can see where the work slows and what it returned, get to close it.

The tools got a decade of investment. The operating model got none. That is the imbalance worth correcting.

_Updated September 13, 2026 to clarify the scope of Figma’s research and distinguish its findings from our interpretation._