In the race for artificial intelligence, the giants have chosen their land: the gross performance of the models. At Anthropic, Openai, Google Deepmind or Microsoft, most of the resources are devoted to the training of ever faster, and more reliable systems. The user experience remains for the moment in the background, the value is still played on the power of the engine, not on the shape of the steering wheel.
And the status quo does not slow down the adoption, because the standard interface, a text box in which we enter a request, followed by an answer, works perfectly. It is understood by all and fits without friction in existing tools. This choice is a classic reflex in technology: the skeuomorphism.
Like the first note-taking applications that imitated a leather notebook or the first e-commerce sites that copied the shelves of a physical store, AI actors reproduce familiar codes of the search engine or online cat. This continuity reassures and accelerates adoption, but it bridles the exploration of new interaction paradigms.
Given from the outside, this distance from the experience design could look like an ideal opportunity for startups. In fact, space is narrower than it seems. A UX innovation that seduces users can be copied and integrated in a few months by the model supplier. And if this innovation only corrects a technical weakness, it can become useless from the next generation. In AI, the rapid improvement of models makes any advantage based on the temporary limitations of an engine.
We already see it on the market. Magicalwhich automates repetitive entry in business applications, will be competed by the Copilot of Microsoft or Google Workspace. Compose aiwhich intelligently completes emails in Gmail, may disappear with the native integration of “Help Me Write” by Google. MerlinChrome extension to question Chatgpt on any page, will lose its appeal when browsers and operating systems will directly integrate these functions. Jasperspecialist in automated marketing editorial staff, finds himself in frontal competition with advanced templates now integrated with Chatgpt, Claude or Canva. Even Otter.apioneer of transcription and summary of meetings, undergoes zoom pressure, Google Meet and Microsoft Teams which now offer these default functions.
For many, the most realistic outcome is not the domination of a market but theacquihirenamely to be bought to integrate their talents and their expertise produced in the roadmap of a major actor. A destiny which, if he guarantees an exit, also act that the main value was not in the interface itself but in the team capable of designing it.
The more demanding but more sustainable alternative is to specialize in vertical with a high entrance barrier. The sectors where regulations, confidentiality or business depth create inertia that protects the interface from rapid copies. In health, legal, regulated finance, energy or defense, value does not only reside in ergonomics but in the integration of expertise, protocols and data that a general actor cannot absorb as quickly. It is in these niches that startups can build a defensible proposal, by becoming not a simple “skin” of AI, but a certified, integrated and essential tool for a profession.
In this equation, the UX is not a free field, the giants of the AI voluntarily allow a scent of conservatism there, a legacy of skeuomorphism, while knowing that they can appropriate at any time an innovation validated by the market, which leaves little room for startups that want to position themselves on the experience.
