The design stages AI did not speed up

A design engagement is a sequence rather than a single task, and generative tools have compressed one part of that sequence while leaving the rest roughly where it was.

Jakob Nielsen set the arithmetic out in April 2026. In a ten step workflow where every step takes the same time, doubling the speed of a single step improves the total by five percent. The other nine steps are unchanged, and the work that used to queue behind the fast step simply queues somewhere else instead.

That is the shape of most design work currently running on generative tools. Concept production collapsed, and the stages on either side of it did not move at all.

Separating what compressed from what did not

Concept production compressed hardest, to the point where wireframes, layout variants, copy drafts and first pass visual directions are now produced in minutes rather than over days. A designer who brought two directions to a review last year can bring eight to the same meeting today.

Discovery did not compress, because participants are still recruited, scheduled and interviewed one at a time, and a fifteen participant study still occupies the same two to three weeks of calendar. The binding constraint there is other people's availability, which no tool reaches.

Alignment did not compress either. A stakeholder group needs the same number of conversations to commit to a direction. Commitment is produced by people arguing in a room, and the number of rooms available in a week has not changed.

Delivery did not compress, since engineering estimates, QA passes and release trains run on clocks that were never the design team's to set.

Four stages, one of which got dramatically faster. The project plan in most organisations was never revised to reflect that, so the schedule still assumes a uniform speedup that did not happen.

Finding where the delay went

The delay did not disappear so much as move to the review, which is where the cost of the extra options is now paid.

A review scheduled for two directions and handed eight becomes a different meeting entirely. The group can see that six of the eight are viable, and nobody in the room can say why one is better than another. The reasoning that used to accumulate while the work was being drawn was never produced, so the options arrived without it.

Meetings like that end one of two ways. A decision is made on taste, and it holds until the first stakeholder who was not in the room sees the work. Or the decision is deferred, another round is commissioned, and the time saved in production is spent twice over in calendar.

The bottleneck moved from producing options to justifying them, which matters because production capacity was never the scarce resource in an enterprise engagement. Agreement was.

Replanning around the new constraint

Three changes are worth making before the next engagement is scoped.

Measure the stages separately, because a single project level velocity figure hides which stage actually moved. Timing discovery, concept production, alignment and delivery as four numbers makes the five percent problem visible on the first project rather than the fourth.

Cap what arrives at a review. Three directions carrying a stated rationale each will decide faster than eight carrying none. The cap does not restrict exploration, which happens before the meeting. What enters the meeting is the subset someone is prepared to argue for.

Move the recovered hours upstream. Hours freed in production are best spent on discovery and alignment, which are the stages now setting the schedule. A team that reinvests them in generating more directions has optimised the step that was already fast.

What this changes about scoping

An engagement priced on production hours is priced against the one input that got cheap. The work that determines whether the project lands is research, framing and the navigation of a stakeholder group toward a decision. That work is unchanged in effort and has become a much larger share of the total.

Proposals written against the old ratio will keep coming in under on production and over on everything else.