In my view, the problem is not typically the intelligence of the models, it’s the user experience (UX).

In their haste to market technological marvels, many AI startups are building brilliant but brittle black-boxes in a vacuum. They are delivering phenomenal intelligence that fundamentally does not fit into existing industry-specific or even generic human workflows. The result is a growing chasm between great technical capability and truly usable software. This organisational friction is leading to a tidal wave of failed projects, a rise in unsecure 'shadow AI' usage, 'agent washing' criticisms, and valid fears of a bubble bursting on the horizon. We are heading straight for the “valley of disillusionment” because the technology is forgetting the most important element – the user.

At Definely, we have found our counter-intuitive path through this noise – we are unapologetically 'AI-second.' We even have a company saying, "The second mouse gets the cheese!"

In 2017, we didn’t set out to build an AI company, we set out to solve a deeply human problem. My co-founder and I were both corporate lawyers at big firms. Being registered blind, access to information has always been challenging; however, this was further amplified when I attempted to navigate complex, multi-referenced legal documents, which is a daily task for most lawyers. We set about trying to build a tool to solve that specific pain point, and quickly realised that what made documents accessible for a lawyer with a disability made them instantly more efficient for all lawyers. The core insight was that the technology had to serve the user’s established workflow, not shatter it.

To truly solve enterprise problems in legacy industries like law, finance, or pharma, you cannot just hire a team of data scientists and hope for the best. Your core product team needs real users – people with very specific, lived experience of the workflows you are looking to support. They are the only ones who understand the unwritten rules, the high-stakes contexts and the psychological burdens of the existing system. When building a product, the first question isn't  "what can the model do?" but "how does the user normally do this task and what are their frustrations?"

The 95% failure rate is a function of companies optimising for their technology rather than for their customers’ adoption. The future of enterprise AI isn't about the model with the largest parameter count or the shiniest ChatGPT wrapper, it’s about the company that nails the last mile of integration. This means understanding not only the sales process but also the lengthy and challenging adoption process.

A six-figure contract is meaningless if the software sits unused. For adoption to stick, the tool must solve a problem in a way that is immediately intuitive and requires minimal behavioural change. It must slot effortlessly into the existing "digital factory" of the enterprise. This requires a profound empathy for the end-user that can only come from having those users embedded in your design process.

The AI gold rush will eventually subside, and the companies that survive will not be the loudest, but rather those that are most deeply integrated into the enterprise ecosystem. For any startup looking to navigate the coming disillusionment, the lesson is clear – stop building a black-box marvel. Instead, focus on building a transparent, user-centric solution. Put the 'U' back in UX, because without a delighted and productive user, your brilliant technology is worth nothing.

For more startup news, check out the other articles on the website, and subscribe to the magazine for free. Listen to The Cereal Entrepreneur podcast for more interviews with entrepreneurs and big-hitters in the startup ecosystem.