A founder scrolls Product Hunt on a Sunday evening. Twelve new SaaS products launched in the last twenty-four hours. The hero sections arrange themselves into a grid that looks oddly uniform: gradient meshes, geometric backgrounds, identical sans-serif typography, dashboard screenshots rendered with the same Lucide icons in the same shade of muted purple. The founder, who has built an MVP this weekend, opens their own product in a tab. It fits in. It is competently designed. It is also indistinguishable.
This is the AI MVP problem nobody is talking about, because every individual MVP looks fine. The problem only appears at scale, in the feed. AI build tools have made design accessible enough that the feed is starting to feel like fifteen hotel lobbies in fifteen different cities, all decorated by the same firm.
Why the Convergence Happens
Three forces are pushing every AI MVP toward the same look. The first is shared defaults: most AI build tools use Tailwind, most pull from shadcn/ui or its lookalikes, most reach for Inter or Geist when nothing else is specified. The second is the training set: the models powering these tools have learned what SaaS products look like by ingesting screenshots of every SaaS product, which produces an aesthetic mean rather than an aesthetic outlier. The third is the prompt itself. “Build me a dashboard” gets you the median dashboard, because nothing in the prompt says what kind of dashboard.
None of this is the tools’ fault. They are doing what they were trained to do: produce a confident, competent result given an underspecified request. The result is good. The result is also indistinguishable from every other good result built the same way.
What Sameness Costs
In a feed, the first job of any product launch is to stop the scroll. A product that looks like the four products before it does not stop the scroll. It contributes to the wash of beige that the viewer compresses into “another SaaS thing.”
The cost compounds at the moment of memory. The viewer might remember “I saw a product about X” but cannot picture it, because the picture in their head is the average of fifteen products that all looked the same. The product has been absorbed by the average.
A homogeneous look also signals that the product was built quickly and casually, even when it wasn’t. The visual sameness implies functional sameness. The viewer extrapolates: if the design is generic, the thinking probably is too. This is unfair, but it is how feeds work.
Three Strategic Moves That Break the Sameness
The fix is not “better design,” because the design itself is competent. The fix is upstream of the design, in the brief.
Pick a visual position before you prompt. “Make it look professional” gives you the median. “Make it look like editorial fashion magazines from the late 1990s, with Helvetica and a lot of negative space” gives you something else entirely. AI build tools can build to a specific aesthetic when one is specified. Most teams skip this step and inherit the default.
Specify what you want to NOT look like. Anti-references are as useful as references. “Don’t make this look like Linear or Notion” gives the build tool a direction to push away from. Without anti-references, the tool drifts back to the median, because the median is what it knows best.
Decide who your visual audience is and what cues they recognise as native. A product for designers needs cues designers recognise: considered typography, Swiss grid, deliberate restraint. A product for engineers needs different cues: terminal aesthetic, monospace, no decoration. A product for executives needs different cues again. Generic AI defaults serve none of these specifically.
These are not technical decisions. They are positioning decisions, made before any AI tool gets touched. The tool then has something specific to build toward.
Escaping the Average
Producing the visual position upstream of the build is the kind of work Zynkex is built for. The page or dashboard or app is the downstream artefact. The reference set, the anti-reference set, and the audience cues are the upstream context that decides whether the build tool optimises against the average or against something specific.
A Zynkex session walks you through where the product wants to sit in visual space, what it wants to be associated with, and what it wants to be distinguished from. What comes out is a Strategy Plan with a prominent Decision Log capturing every call made and why, plus a Build Brief that gives the build tool a specific aesthetic destination, not a default one. The execution still depends on the build tool’s capability and on your design taste. What changes is that the build tool is no longer guessing.
The default of AI design is the average of all AI design. The average is the safest place to be and the easiest place to disappear. The work to escape the average is upstream of the prompt, and that is where the difference between a product that fits in and a product that gets remembered is decided.
