When a candidate uses AI to complete an assignment, the immediate reaction is often that the assessment has somehow been compromised. The assumption is that if ChatGPT or Claude helped create the work, we are no longer seeing the candidate's real ability.
I think that is the wrong question.
The more useful question is: what are we actually trying to measure?
This matters particularly when hiring junior or early-career designers. If someone has ten or twelve years of experience, there is usually a body of historical work, real projects, decisions, outcomes and references that give you other ways of understanding their capability. An assignment is a different proposition when it is one of the few substantial pieces of evidence you have about a less experienced candidate.
And AI changes what that assignment can tell us.
Know what is a Given and what is to be judged
Imagine asking two people to write the same statement on a paper.
The first person uses a brush and ink and creates the lettering by hand. The second person uses a computer and prints it.
Wuld not judge both people on the shape of the individual letters?
For the first person, the formation of those letters is part of the skill being demonstrated. For the second, it isn't. The letterforms came from the typeface. What you can judge is the choice of typeface, the composition, the spacing, the hierarchy and how effectively the person used the available resource.
AI creates a similar situation for design.
Give ChatGPT or Claude a reasonably well-defined design problem and it can produce a surprisingly competent starting point. It can suggest layouts, typography, colours, copy, responsive structures and many of the other elements that previously required a designer to produce from scratch.
That does not mean there is no design skill left to evaluate.
It means we need to be more precise about where that skill is.
Manual calculations are no longer a desired skill
This is not actually a new problem.
When I was in school, we used logarithmic tables to solve certain problems in exams. The tables gave us something that would otherwise have taken considerable manual effort. Nobody was particularly interested in whether we could reproduce the underlying calculations from memory. The point was whether we could use the available resource to solve the problem.
A calculator in an accountancy examination works in much the same way. If the objective is to understand whether someone can manage and interpret a balance sheet, testing whether they can multiply and add numbers without assistance is not necessarily useful.
The tool removes one layer of mechanical effort so that you can examine something more meaningful.
AI can be treated in the same way.
If a designer is going to use AI as part of their everyday work, then an assignment that deliberately removes AI may be testing for a version of the job that does not exist.
If the tool is part of the job, the ability to work with the tool is part of the capability you are hiring for.
There are two different kinds of assessment
I love watching Forged in Fire with my son.
The final round of the competition specify the dimensions, materials and characteristics the finished weapon must have. The contestants go back to their own forges. They may have different equipment and different techniques available to them and they are not told to do things in a specific way. What matters is whether they can produce the required result in the given timeframe.
An interesting variation is when the competition deliberately controls the environment. Sometimes (at least once) contestants are placed in a forge and are required to start the coal forge themselves, then that becomes part of the challenge. The conditions have been intentionally designed to test something specific and they are similar to every smith, no one can sneak in a gas powered forge in there.
Similarly, there can be perfectly valid reasons to conduct a no-AI assessment.
If so, create the conditions that allow you to measure that fairly.
Bring the candidates into your office. Give them the same computer, the same software, the same information, the same amount of time and the same constraints. Remove internet access or AI access if that is what you need to control.
That is a controlled assessment.
In other cases restricting the tools they will use in the real job may make the assessment less representative rather than more rigorous.
Control what matters
This leads to a simple principle for designing hiring assignments:
Before deciding what tools a candidate can use, decide what capability you are trying to measure.
If you are hiring someone who will spend their days using AI, Figma, design systems, research tools and other resources, there is little value in pretending those resources do not exist.
Instead, define the things that should be fixed.
The problem might be fixed. The target audience might be fixed. The business objective might be fixed. Certain technical or accessibility constraints might be fixed. The expected outcome might be fixed.
How the candidate gets there can be more open.
That gives you a much more realistic opportunity to see what they do with the resources available to them.
The value is in what happens above the baseline
Imagine giving five candidates the same problem and giving all five access to AI.
You may get five reasonably polished designs.
That is not necessarily a problem.
In fact, it can be useful. You have established a common baseline.
Now you can look at what each candidate did beyond that baseline.
Did they understand the problem properly? Did they question the assumptions? Did they identify something the initial AI-generated solution missed? Did they make sensible trade-offs? Did they simplify something that had become unnecessarily complicated? Did they recognise that a visually attractive solution did not necessarily solve the user's problem?
Did they use AI as a shortcut to an answer, or as a way of exploring possibilities before applying their own judgment?
Those differences may be much more useful in predicting how the candidate will work than whether they can independently produce a polished first screen.
This is particularly important with junior designers. They may not yet have a long portfolio of shipped work from which you can judge their professional experience. The assignment therefore needs to reveal how they approach a problem, not simply how quickly they can manufacture an artefact.
Don't make AI the enemy of the assessment
There is a temptation to spend a lot of energy trying to detect whether a candidate used AI.
But that can turn the hiring process into a policing exercise.
The more useful approach is to make the role of AI explicit.
If AI is allowed, allow it for everyone. If it is not allowed, create a controlled environment where that restriction can actually be enforced. In either case, make sure the restriction exists for a reason.
The mistake is to assume that an assignment is better simply because it is harder.
A harder test is not necessarily a better test.
If you ask a candidate to work for eight hours without tools they will have access to every day once they join, you may simply be measuring their ability to tolerate an artificial constraint.
If, instead, you give them the same resources they will have on the job and ask them to solve a meaningful problem, you may get a much better view of how they actually work.
The question to ask before the assignment
The next time you create a design assignment, don't begin by asking whether candidates should be allowed to use AI.
Start one step earlier.
Ask yourself: what am I trying to learn about this person?
If you want to know whether they can draw a screen from scratch, then test that.
If you want to know whether they can solve a user problem, make decisions and produce a good outcome using the tools available to them, then test that instead.
If you want to know how they think without assistance, create a controlled environment.
If you want to know how they will work in your organisation, give them an environment that resembles the one they will actually encounter.
AI has not necessarily made design assignments impossible.
It has simply made some of our old assumptions about what those assignments measure more visible.
The goal of a hiring assignment was never to prove that a candidate could work without tools.
The goal was to learn something meaningful about the person you were considering hiring.
That goal hasn't changed.
Perhaps the assignment needs to.
