The portfolio worked because it was hard to fake

The design portfolio was a useful hiring signal because of what it cost to produce. Building a case study for a complex product took weeks. You had to have actually shipped the thing, or at least been close enough to the decisions to reconstruct them credibly. That cost was a filter. It kept the signal honest. If someone walked you through a portfolio of five deep case studies, you knew something about their history, even before they opened their mouth.

AI collapsed that cost. The filter broke. That's constraint inversion. The portfolio's value came from the constraint it imposed (execution is expensive, so only people who did the work can show the work). Remove the constraint and the artifact becomes noise. The question: what constraint, if restored, would make the signal honest again?

The response made things worse

The industry response has been to double down on the broken proxy. More polish. New sections explaining "how I used AI in my process." Figma's own 2026 survey found that 73% of hiring managers now want candidates to demonstrate AI tool proficiency. That is a demand for more performance, not more signal.

The advice circuit followed. Document your process. Add a "tools used" section. All of this assumes the portfolio is still the right container and just needs updating. None of it asks whether the container itself lost its filtering power.

Jakob Nielsen's research across 2.26 million freelance contracts showed what happens when competence gets commoditized. After ChatGPT launched, the signals freelancers spent careers building (credentials, portfolios, five-star ratings earned through real transactions) lost hiring weight in AI-exposed fields.

Nine controlled experiments showed why: AI lifts the least-skilled workers the most, narrowing gaps until competent work floods the market. Once competence is abundant, it becomes a commodity.

Design portfolios hit the same wall. As Everyday UX put it this year, "the 'good enough' portfolio has become a commodity." Having shipped projects and deep case studies is now the prerequisite, not the finish line. The baseline moved, and everything built on top of it (the storytelling, the Dribbble-ready screens) floats on a foundation that no longer differentiates.

Who restored the constraint

Some hiring teams figured this out. The Fountain Institute reported that a growing number of first-round interviews skip the portfolio walkthrough entirely and open Figma instead.

That's constraint inversion done right. The old constraint (execution is expensive) disappeared. The new constraint (think under pressure, in real time, with no preparation) restores the signal by testing something AI cannot yet do: decomposition on the spot. You hand someone an ambiguous problem and watch how they break it apart. No prep. No polished narrative to fall back on.

The decomposition was always the job. A senior designer's value lives in deciding which screens to produce, in what order, with what tradeoffs. AI made that impossible to hide by making everything else trivially easy.

Pawel Klasa put it this way earlier this year: "AI can fake a portfolio in an afternoon. It cannot fake twelve months of public thinking." The portfolio shows finished products. The live exercise shows decision-making. One of those still works as a signal.

The longer hiring pipeline is a symptom

Tom Scott noted that hiring managers now face "loads of talent that have AI CVs or even portfolios, so the hiring process gets even longer than it was." That lengthening follows from broken signal. When the filter does not filter, you compensate with more stages and more gut-feel evaluation. The process gets expensive and slow because the cheap proxy stopped working.

Adding "show your AI process" to the portfolio does not fix this. It adds another performable layer on top of a performable artifact. The signal stays broken because the underlying constraint (this was hard to produce, so production implies competence) is still gone.

The lens

Constraint inversion is one question: when the answer is unclear, what constraint, if added, would make it obvious? The portfolio lost its constraint. The teams that added a new one (live exercises, observable decision-making over time) got signal back. The teams that kept optimizing the old artifact are sorting through noise and wondering why hiring keeps getting slower.

If you are a designer asking what replaces the portfolio, the answer is a different constraint. One you cannot generate in an afternoon.

Sol's diagnosis is clean. The portfolio was a proxy for having done the work. The cost of producing it was the filter. AI killed the cost. The filter broke. Constraint inversion says add a new one. Live design exercises add a new one. All correct.

The part Sol skipped: live exercises test decomposition in a vacuum. They test whether someone can break an ambiguous problem apart under time pressure in a room with strangers watching. That is a real skill, but not the one that determines whether a designer ships something useful.

When I hired for Merdeka Mengajar, I did not care how someone decomposed a whiteboard prompt. I cared whether they had shipped anything that worked for a teacher on a cheap Android in a district where the internet cuts out after lunch. You cannot perform that in forty-five minutes. You accumulate it across years of contact with a context that does not exist in an interview room.

Sol quotes Pawel Klasa: "AI can fake a portfolio in an afternoon. It cannot fake twelve months of public thinking." That is the stronger signal, and Sol identified it but did not follow it. The live exercise is a thirty-minute replacement for a broken artifact. Twelve months of public thinking is a structural one. Public commits and decisions. Public reasoning you cannot retcon because it is timestamped.

The live exercise also strips context. A portfolio, even a fakeable one, at least showed where someone worked and who they designed for. A Figma exercise in a first-round interview tells you whether the candidate can think on the spot. It tells you nothing about whether they have ever sat across from a user who does not share their assumptions. Cultural literacy shows up in choices made over months, for a specific population, under real constraints, not in a timed exercise.

Sol is right that the portfolio container is broken. The replacement should be a longer trail, not a better test. Ship things and write about the decisions publicly. Do it long enough that the body of work becomes the constraint. That is what AI cannot fake in an afternoon or a year.