There’s an ad doing the rounds at the moment. The story goes something like: everyone in the workplace has built an app. It’s slick, it’s optimistic, and it’s misleading in exactly the way that tech marketing tends to be.
Yes, the tools are extraordinary. You can describe a product in plain English and watch something functional-looking appear in minutes. What the ad doesn’t show you is what happens next — when that something functional turns out to solve the wrong problem, confuse its users, fall apart under real conditions, or simply never get used at all.
And thinking — real, structured, research-backed product thinking — is precisely what us Product and UX designers have spent our careers learning to do.
The numbers are striking. According to a 2026 RAND Corporation analysis, 80.3% of AI projects fail to deliver their intended business value. A third are abandoned before reaching production. Nearly another third ship — and still fail to solve anything meaningful.
The single biggest predictor of failure? 73% of failed AI projects had no agreed definition of success before the project started. Projects that defined clear, measurable goals upfront achieved a 54% success rate. Those that didn’t: 12%.
That gap — 12% versus 54% — is not a technology problem. It is a thinking problem. It is a research problem. It is a brief problem.
Meanwhile, 43% of AI-generated code requires manual debugging in production even after passing testing, and 45% contains security vulnerabilities. The tools are producing output at speed. The output often isn’t good.
The market is not being flooded with great products. It’s being flooded with built products. Those are very different things.
Think about what happened when desktop publishing arrived in the 1980s. Suddenly anyone could lay out a page. What followed was a decade of truly terrible design — because access to the tool was never the scarce resource. Judgement was.
The same pattern is playing out now, at speed and at scale. AI has democratised buildability. It has not democratised the ability to identify a real problem worth solving, to understand the people who have that problem, or to structure a solution in a way that actually makes sense to use.
Those capabilities belong to seasoned designers.
This is the shift that most people are missing. The constraint in product development has moved upstream. It’s no longer “can we build this?” — it’s “should we build this, and do we understand it well enough to brief it properly?”
Consider what it actually takes to get a good outcome from an AI build tool. You need to define the problem with precision. You need to understand your users well enough to make decisions on their behalf. You need to map out the flows, the edge cases, the states, the logic. You need to know what “good” looks like before you can recognise whether you’ve achieved it.
In other words: you need to create a proper brief with good research to refine the concept before the prototyping/build can begin.
Most people don’t know how to do that. Many don’t know it’s required. They describe what they want at a surface level, generate something that broadly resembles it, and ship the result — confused about why it doesn’t perform, why users drop off, why the thing that looked right doesn’t feel right in practice.
This is the part the ad won’t show you.
A genuinely useful product brief is not a paragraph. It’s the distillation of user research, competitive context, business constraints, defined success metrics, user flows, edge cases, and a clear articulation of what this product is not trying to do. It requires the ability to hold complexity, resolve contradictions, and translate messy human behaviour into clear design decisions.
But more than anything, it requires a real human, living in the real world who understands real human problems and can create a useful product concept from that - i.e. insight.
That skill takes years to develop. It cannot be generated. It has to be applied by someone who has done the work.
When I recently built Ticklists.com with AI assistance, the quality of what came out was directly proportional to the depth of the brief that went in. I started with a real-world user problem - sharing a shopping list with my partner so that he can see everything to buy in the supermarket in EXACTLY THE RIGHT ORDER to cut down his shopping time and frustration - and this works for any store he chooses to go into. Same real-time synced list, different stores, perfect order.
Any ambiguity in the input can become a problem in the product. Every assumption I didn't examine surfaced as a gap in the experience. The AI was a capable builder. I had to be the architect.
That is not a temporary state of affairs. It is the model.
There is a useful distinction between building and designing that tends to get lost in the current noise.
Building is the creation of something functional. Designing is the process of understanding a problem deeply enough to solve it in a way that is useful, usable, and worth using. These are not the same activity, and one does not substitute for the other.
What AI has done is make building cheap and fast. It has done nothing to make the design process less necessary. If anything, it has made the consequences of skipping it more visible — because now anyone can ship quickly, the market fills up faster with things that don’t work, and the contrast between a well-designed product and a generated one becomes harder to ignore.
Our ability to conduct research, synthesise it, define a concept, map an information architecture, and write a brief that actually holds — that is design. That is what determines whether a product succeeds, and is highly valuable.
UX research, far from being made redundant by AI, is becoming more strategically important. The number of organisations where research is considered essential to all levels of business strategy nearly tripled in a single year — from 8% in 2025 to 22% in 2026. Demand for research increased from 55% to 66% in the same period.
AI tools are accelerating the analysis phase — cutting qualitative analysis time by up to 80% in some workflows. But someone still has to know what questions to ask. Someone still has to decide which patterns matter and which are noise. Someone still has to translate insight into direction.
That is Designers' skill. And it is being asked for more, not less.
The instinct that AI tools encourage is speed. Generate, iterate, ship. And there is genuine value in that — prototyping has never been cheaper, and the ability to test ideas quickly is real.
But speed without direction is just movement. The designers who will thrive in this environment are the ones who understand that the thinking cannot be skipped — it can only be done badly and invisibly, until the product fails and nobody quite knows why.
Design thinking — the structured process of understanding, defining, ideating, prototyping, and testing — is not a relic of a slower era. It is the process that turns a build into a product. It is what separates the 54% of AI projects that succeed from the 46% that don’t.
The bottleneck has shifted from execution to thinking. That is a remarkable development for anyone who has spent years being told their process is too slow, too costly, or too hard to justify.
It was never any of those things. The market just needed a moment to find out what happens without it.
The skills that have been hardest to quantify and easiest to cut are now the ones that determine whether an AI-era product works or not.
Problem framing — defining what is actually worth solving, for whom, and why. AI cannot validate whether a problem is real. Most people building with AI skip this entirely.
Information architecture — the structure of a product is invisible when done well and catastrophic when it’s wrong. Non-designers don’t know this is a discipline until they’ve shipped something that confuses everyone.
Research synthesis — the ability to absorb messy, contradictory human input and distil it into clear product decisions. AI can speed up the analysis; someone still has to direct it.
The brief — precise, structured, detailed enough that the build actually solves what it’s meant to. This is a craft. It takes practice. Most people have never had to write one because they could just hand a vague spec to a developer and negotiate meaning over months of build. Now they have to be the brief themselves, and they are not ready for it.
The power tool is available to everyone. The trained eye that knows what to build with it — and how to brief it properly before picking it up — is still rare.
___________________________________
References - Statistics:
1. AI project failure rates (80.3% fail, 73% had no success definition, 54% vs 12% success rate)RAND Corporation analysis, reported via Pertama Partners and Folio3 AI, 2026. Global scope, primarily enterprise AI initiatives.
2. AI-generated code requires debugging in production (43%)CodeRabbit Developer Survey, December 2025, reported via VentureBeat. Global scope.
3. AI-generated code security vulnerabilities (45%)Veracode State of Software Security, 2025. Primarily US and global enterprise data.
4. UX research strategic importance (8% → 22% essential to business strategy; demand 55% → 66%)Maze Future of User Research Report, 2026. Global scope.
5. AI cuts qualitative analysis time by up to 80%Maze UX Statistics, 2026. Global scope.
References - Links:
1. AI project failure rates (RAND / 80.3%, 73%, 54% vs 12%)
2. AI-generated code debugging (43%)
3. AI code security vulnerabilities (Veracode 45%)
4. UX research strategic importance (8% → 22%, 55% → 66%)
5. AI cuts qualitative analysis time 80%
Learn by watching me apply key design principles
1 email per week with videos & free resources
Unsubscribe anytime. See a sample
X
All the Visual Design Principles in one book with examples. Learn, refresh or hone your design skills to create stunning designs - Available as e-book for €7.99
Leave your details to hear when it's ready!
