AI Didn't Flatten Your Product. Your Spreadsheet Did.

The cheap-and-measurable option just got cheaper. Defending the unmeasurable is now a senior leadership job, not a craft argument.

Two pieces published in the last fortnight should land like a tuning fork on anyone running a design, product, or program function right now.

The first is Yann Caloghiris writing in The Drum on May 21, walking through a senior UX expert's stress-test of AI tools against a real design workflow. The finding: AI handles roughly 80% of the design process competently — flows, structure, first-pass prototype, the lot. The remaining 20% — the interaction feedback that makes a user feel understood, the visual language that communicates brand without words, the creative direction that signals someone actually cared — is where brand experience actually lives. And most organisations are now shipping the 80% and calling it done. Not because they evaluated the trade-off and accepted it. Because the tool produced something plausible enough that the question never got asked.

The second is Vadym Grin's Eidos Design dispatch on May 25, naming what the rest of us are already feeling: AI-generated products wear a uniform. Purple gradients. Thick sans-serifs. Blobs and abstract backgrounds. Rounded corners stacked around more rounded corners. Pillow buttons. Soft shadows on every surface. Scroll any Product Hunt page this month, Grin writes, and the fingerprints repeat. Designer Michal Malewicz has put a name to the resistance — Slopless, an anti-sameness manifesto — and a small, real movement is gathering behind it.

The temptation is to read both pieces as a craft argument. Designers worried about taste. Aesthetes complaining about gradients. I think that's almost exactly wrong.

This is not an aesthetic problem. It is an economics problem with an aesthetic surface.

The 80% Trap.

Caloghiris and his colleague Natalie Levy-Acosta name the mechanic precisely. AI gets you 80% of the way there in roughly 20% of the time it used to take. That is a real efficiency gain. It would be foolish to ignore it. But the remaining 20% — the brand specificity, the creative direction, the human judgment that decides whether something feels like us — has not gotten cheaper. It has gotten dramatically more expensive in relative terms, because the rest of the work just collapsed in cost.

The number that has stuck with me from the piece is from Figma's design survey: 78% of professionals say AI tools meaningfully accelerate their workflows. Only 58% say AI actually improves the quality of the output. That 20-point gap is the whole story. It is the gap between cheaper and better. It is the gap that, if no one in your organisation explicitly defends, simply closes by getting cheaper.

I have watched some version of this transition arrive at every interesting moment. The cheaper option lands. It hits a threshold of "good enough." Procurement notices. The next time someone tries to justify the more expensive path, the conversation gets shorter. Eventually, the more expensive path stops being on the table at all, and nobody quite remembers when that decision got made — because it wasn't a decision. It was an erosion.

Goodhart Was a Design Ops Manager.

The deeper engine here is Goodhart's Law in business clothes: when a measurable proxy stands in for the thing you actually care about, the proxy eventually replaces the thing.

In design, the proxies are Lighthouse scores, conversion rate, time to first meaningful paint. Real. Useful. Not the thing.

In product, they are NPS, DAU, retention curves, A/B-tested variant lifts. Real. Useful. Not the thing.

In design ops, they are design-system adoption percentage, Figma file counts, ticket throughput, designer-to-engineer ratios. Real. Useful. Not the thing.

In program management, they are velocity, burndown, on-time delivery. Real. Useful. Not the thing.

The thing — in every case — is some compound of considered judgment, brand voice, durable trust, taste, the user's quiet sense that someone on the other side of the screen actually thought about them. These compounds are not measurable in any way a CFO will accept on Monday morning, which is exactly why they lose every quarter to whatever is. Douglas Brundage's recent essay on AI copywriting put the dynamic bluntly: JPMorgan Chase didn't sign Persado's five-year deal because the copy was that good. They signed it because clicks went up and cost went down. Persuasion was never on the spreadsheet.

What AI changes is not the existence of this trade-off. The trade-off has always been there. What AI changes is the price. By collapsing the cost of the measurable option toward zero, it makes the relative cost of defending the unmeasurable look enormous. The math gets brutal in a way it wasn't five years ago.

The Slop Rebellion is the polite name for a much harder conversation: what are you willing to be more expensive for?

You Can't Defend the Unmeasurable From the Middle.

Here is the part that rarely gets said out loud. Designers don't own the P&L. Most PMs are compensated against exactly the metrics that are eating their craft. Program leaders can name dependencies but rarely redefine value. The people who feel the loss most acutely are the people with the least authority to stop it.

This is why "designers should care more about taste" is a worse-than-useless framing. They already do. What they don't have is the political cover to slow a release for something they can't put on a slide.

Defending the unmeasurable is, in operational terms, a senior leadership move. It is a CEO, CPO, or CDO saying we are going to under-optimise on this measurable in service of this less-measurable thing, and meaning it after the next board cycle, and the one after that. It is the discipline that brand strategists have always practiced. I went through a multi-week mentorship with Musa Tariq this spring, and the line that has stuck with me most is also the simplest one in his canon: a brand is either interesting or invisible. There is no middle. The "interesting" path is, definitionally, the one that resists optimisation toward the average — which is to say, the one that nobody can fully justify before the fact.

Brand strategy and product strategy are converging here. The argument Caloghiris is making about UX in The Drum is the same argument Musa makes about brand: the average is the death zone, and the average is exactly what optimisation-by-proxy produces.

What Design Ops and PM Can Actually Do.

I want to be useful here, not just diagnostic. If you sit in the middle of an org — design ops, program management, product ops — there is a move available to you that does not require you to win the philosophical argument with finance.

Get in the business of making the unmeasurable legible without making it measurable.

Legibility buys you runway. Measurement kills the thing it tries to capture.

In practice: write the line. In your release notes, your readout decks, your status updates, name what you protected and why.

We slowed down because we didn't want to ship something we'd be embarrassed by in three years.

We picked the harder direction because the easier one was one more frictionless flow nobody would remember.

We rejected the AI-generated first pass because it was confidently 80% of a product, and we needed the other 20%.

We added one weird detail on purpose, because the version without it tested fine and felt like nothing.

None of these statements ask the spreadsheet to grow a soul. They give the senior leader who has to defend craft to the board a sentence they can actually use. They turn craft from a vibe into a position. They are, themselves, small acts of authorship.

The risk worth naming out loud is that defending craft can also be a cope. It can mean defending slowness, ego, or aesthetic preference dressed up as principle. The version that holds up is craft-in-service-of-the-user, or craft-in-service-of-durability — not craft-in-service-of-the-designer's-portfolio. The leaders who actually win this argument inside their companies are the ones who can tell those two apart. The Slopless manifesto is interesting precisely because it makes that distinction explicit: it is not anti-AI. It is anti-handover. It asks designers to keep authorship of the decisions that AI cannot, and should not, be making for them.

The Tension Worth Sitting With.

Here is the uncomfortable bit for our discipline. Design ops, product ops, and program management as functions exist, in part, because we made design and engineering legible to the business. We professionalised by adopting metrics. We built the dashboards. We translated craft into capacity and velocity and adoption percentages so that the CFO would fund the team.

That work was necessary. It is also exactly the work that is now being used to justify replacing the craft we were translating.

The next move for our discipline is not to abandon legibility. It is to use the credibility legibility bought us to defend the things legibility can't capture. To stop being only the function that makes design measurable, and start being the function that names — out loud, in the meetings that matter — what is being lost when measurement becomes the whole game.

The floor of competence is rising. AI is going to keep raising it. That is mostly good news, including for the people reading this. The question is not whether your team can hit the new floor. The question is whether anyone in your organisation is being paid to care about the ceiling.

Make sure that somebody is. Ideally, somebody senior. Ideally, you have already given them the sentence.

Further reading: AI gives us the prototype. It doesn't give us the brand — Yann Caloghiris, The Drum, May 21 2026 · The Slop Rebellion — Vadym Grin, Eidos Design, May 25 2026 · Slopless manifesto — Michal Malewicz · Why Can't AI Write Copy? — Douglas Brundage, Enfant Terrible · Science x Story — Musa Tariq & Tim Jones