Why this work stays close to people
Set and exhibit design starts with a story and ends in a physical room. A designer reads a script or a curator’s brief, decides what a visitor should feel on arrival, then works out how plywood, light, paint, and traffic flow can deliver it. Software can draft a version of that room. It cannot stand in the half-built gallery at 7 a.m. and decide the case is six inches too tall.
Two parts of the job carry most of that weight. The first is translating a brief into a spatial concept that fits a real venue, with its sightlines, load limits, doorways, and fire exits. The second is supervising the build: checking fabrication against the drawings, approving substitutions when a material falls through, and adjusting on site when the lighting rig changes what the color looks like. Both involve judgment under pressure, with other people in the room.
The scale of the job matters too. US employment is small, about 10,630 jobs, with median pay of $75,240 a year and projected employment change of 1.4% from 2025 to 2035 (BLS, 2025). A small field with long project cycles tends to change slowly, because each theater, museum, or trade-show client buys design as a one-off service rather than a repeatable output. Our coverage measure, which estimates the share of task time AI can handle today, reads 26 out of 100; you can read how that is built on the coverage method page.
What AI does, assists with, and leaves alone
AI is already useful on the flat, early, and repetitive parts. Generating mood boards and concept images from a written brief is fast. So is producing variant layouts, preliminary drawings, and first-pass material or prop lists from a model. Of the task time AI touches in this job, 0% is work it can carry on its own.
A larger part of that exposure is assistance rather than handover: 44%. Here a designer still drives. Estimating budgets and schedules, rendering a set in 3D for a director or curator, and writing up specifications for a fabrication shop all go faster with tools, but the numbers and choices get checked by the person whose name is on the drawing.
The remainder stays with people: 56% of task time. That is the site visit and venue survey, the back-and-forth with directors, producers, and curators about what the exhibit is really saying, and the supervision of carpenters, painters, and installers through load-in. Robotics is not the obstacle either; only a small slice of this job is physical machine work, and our robotics tier for it is “none needed”. The holdup is coordination and taste, not hardware.
What the evidence actually shows
There is no direct test of AI against working set and exhibit designers yet. Our evidence grade for quality parity is D, and a grade at that level means we publish no parity number at all. We will not guess one. How grading works is set out on the quality parity method page.
What would settle it is fairly specific. A blind review where experienced designers and an AI system each produce a concept, layout, and materials plan for the same venue brief, judged by producers or curators who do not know the source. Or a measured record of built projects: how often AI-generated drawings survived fabrication review without rework, and how site changes compared. Until something like that exists, the honest answer is that the tools are demonstrably good at images and drafts, and untested on the full job.
Because the parity evidence is thin, the headline figure on this page leans on the task mix and the blockers rather than on any single study. That figure is 73 out of 100 (higher is safer), and the full approach is on the methodology page.
The timing, and what could move it
Most likely between 2037 and 2052 (8 in 10 of our scenarios). The replacement-year method page explains how that window is produced.
Two things could pull it earlier. Tool spend is tiny next to a design fee, as the cost panel on this page shows, so a studio can test AI on concepts without a budget case. And if generated drawings start passing fabrication review with little rework, small venues may buy fewer concept hours per project.
Two things hold it back. Buildings and venues are stubbornly specific: every load limit, sightline, and accessibility route has to be verified in person, and liability for a bad call sits with a named professional. Client work is also relational. Theaters, museums, and exhibit houses hire designers they have worked with, and that trust does not transfer to a tool.
How to stay needed in exhibit and set design
Lean into the tasks that stay with people. Run the venue survey yourself and own the constraints no model knows. Keep control of the conversation with directors and curators about meaning, not just look. Stay on site through fabrication and load-in, where most of the real decisions get made.
Two skills pay for themselves. One is technical fluency with 3D and drafting tools, including the generative features, so you produce options faster than you did last year. The other is budget and schedule command: a designer who can price a change in the room is hard to work around.
What to do: keep a record of site problems you caught before the build, because that is the part of the job nobody can download.
If you are weighing adjacent paths, the closest work sits with Interior Designers, Commercial and Industrial Designers, and Merchandise Displayers and Window Trimmers. The wider art and design workers family page shows how the scores spread across design roles, and the arts and entertainment sector page covers the venues that employ most of them. You can also put two of these roles side by side on the compare tool, or see where design work lands on our list of the safest jobs from AI.