A Portfolio with AI Projects: What to Show and How

What do reviewers look at in an AI project?

There is no reliable survey that measures this, so what follows comes from interview guides, not statistics. TechInterview’s guide to projects “built with AI” (published in May 2026 and updated in July) is written for technical candidates, but several of its ideas carry over to design. According to the guide, reviewers look first at the README, then at the repository structure, the evidence of evaluation, the candor about tradeoffs and the deployed product, if there is one.

The guide offers a simple test: if you could swap the AI for a regular expression or a lookup table and nothing changes, that AI does not count. The model has to do real work, meaning its judgment or generation matters. In design the equivalent question is this: if your case would work the same with a fixed list of options, it proves nothing about AI.

Fueler, in an October 2026 post, says it from the design side: a portfolio that only shows polished images is becoming less useful, and you should show how you think.

What should each case study show?

Combining the TechInterview and Fueler guides, an AI case study should answer this, in this order:

1. What was the problem, and who was it for? Problem first, without mentioning AI yet.
2. Where did AI enter, and why there? Which step changed and which steps stayed the same.
3. Which decisions did you make? What context you gave the model, how much you controlled the output, how you brought the response into the interface, which real constraints you had.
4. How did you evaluate it? A set of cases, a usability test or a baseline metric.
5. What failed? One case where the AI got it wrong and how you corrected it. TechInterview considers candor about tradeoffs a strong indicator.
6. What changed, and what comes next? The outcome and two or three concrete steps to improve it. For a personal project, Fueler suggests explaining what you learned or what you changed after testing it.

How do you show AI use without making it the star?

Fueler sums it up: the goal is not to hide AI but to show that you directed the process instead of accepting the first output. Muzli, a UX portfolio guide from March 2026, asks you to document AI tool use in your case studies, and not to treat it as a gimmick.

In practice:

• Add a short block to each case: where you used AI, what it returned and what you changed.
• Show decisions, not only results. Muzli suggests answering what you tried first and what did not work.
• Show iterations and discarded directions when they help explain the process.
• Do not try to learn or show every new tool; Fueler advises against it.

How many projects, and how deep?

The sources disagree, and it is better to say so than to invent a rule:

A reasonable rule: for a design role, three complete case studies; if you also show code, one deep project that can be run. According to TechInterview, do not delete old projects if they were strong, but add a recent one before a job search.

How do you evaluate an AI project without a research team?

TechInterview states that even 20 hand-curated examples make a meaningful evaluation set. Its advice, adapted to design:

1. Gather about 20 real cases, including edge cases and cases where the model already failed.
2. Write down why you chose each one.
3. Measure something specific: accuracy, whether the answer is grounded in the source, formatting.
4. Show the methodology, not just the score.
5. When you have numbers, give them context. Muzli asks for numbers with context; without hard metrics, it proposes qualitative signals such as stakeholder feedback.

And make it testable. TechInterview says a live URL or a one-command run lets a reviewer try the project in a minute, and that a project that cannot be run is read as a project that does not work.

When should you NOT put AI in your portfolio?

When the project does not need AI to solve the problem. A portfolio that adds AI to everything reads as a list of tools. If AI did not improve the result, say so and show what you learned. To see what is changing in entry-level work, read being a junior designer in the AI era and the design engineer career. If you need prototypes for your cases, start with prototyping with AI.

Example

Illustrative example: a designer redesigns the sign-up form of a health service and uses a model to suggest fields depending on the type of appointment. In the case study she shows the baseline (four minutes and five empty fields on average), the two versions she tested, twenty cases with the result for each, and one case where the model suggested a field that should not have appeared. The fix was a fixed rule, not a longer prompt. All figures are invented to illustrate the structure.

Common mistakes

• Showing only the final screens and hiding the process.
• Presenting AI as magic, with no limits or failures.
• Opening a case with a tool logo instead of the problem.
• Publishing projects nobody can try.
• Claiming savings without a baseline.
• Including projects with no decisions of your own, which are just copied prompts.

Sources

• TechInterview: The AI Portfolio, What “Built With AI” Means in 2026 Interviews
• Fueler: Will AI Replace Junior Designers? What Hiring Data Shows Right Now
• Muzli: How to Build a UX Portfolio That Actually Gets You Hired (2026)