The hidden design flaw in enterprise AI
The majority of C-suite executives say adopting AI is tearing their company apart, while eight-in-ten organisations are struggling with adoption despite record investment. But according to Niklas Mortensen, chief design officer for Europe at Designit, the problem isn’t the technology. It’s what happens when every employee gets their own AI-enabled way of working, and the organisation doesn’t redesign around it.
Ask ten people in the same organisation how they use AI and you’ll get ten different answers. Different prompts. Different workflows. Different ways of approaching the same challenge.
That’s hardly surprising since every major AI platform is designed to learn how individuals work. The more you use it, the more it adapts to your habits, your preferred ways of framing problems and the kinds of answers you’ll find useful. That’s what makes AI such a powerful tool.
But for enterprise AI, that’s an issue: when ten people in the same team are each working with their own personalised AI context, you no longer have one AI-augmented organisation but ten slightly different ones operating under the same roof. In short, the more AI adapts to each of us as individuals, the harder it becomes to maintain a shared way of thinking across an organisation.
A coherence problem
Asana recently described many enterprise AI products as ‘single-player’, a description clearly built to sell their own agents, but accurate all the same about where most enterprise AI still sits. Anthropic has started building toward something different with shared Projects, a workspace where a team can pool documents and context so everyone’s chats draw on the same material, rather than each person building up their own private context from scratch. But shared context isn’t the same as shared thinking: the reasoning still happens in separate, private conversations, and the category as a whole still treats individual productivity as the primary measure of success.
Research from BetterUp’s Kate Niederhoffer backs this up from another angle: AI has decoupled effort from quality. You can no longer tell how much thought went into something by looking at it, and it’s already showing up as measurable erosion in trust between colleagues.
Interestingly, the organisations buying these tools evaluate them in much the same way. Procurement measures adoption, productivity and return on investment. Of course those things matter, and AI is already delivering meaningful improvements here.
Very few ask what these systems might be doing to collective thinking.
Many organisations succeed because people develop a shared understanding of the problems they’re trying to solve. They develop common language, shared judgement and collective ways of making decisions. That’s much harder to measure.
On the face of it, nothing appears broken - in fact, productivity improves. But what the organisation can’t see happening, because it’s happening by design, is the gradual loss of something much more difficult to track: coherence.
The upshot is that both sides are optimising for the individual rather than designing for the organisation.
When two colleagues used to work something out, they left a trail, a shared doc, a meeting, a thread other people could see and build on. When they each work it out with their own AI, that reasoning happens in a private chat only one person can see. The output looks like progress.
What’s missing is the paper trail that used to let an organisation catch itself drifting. The coherence problem isn’t simply emerging inside organisations; it’s built into the way today’s enterprise AI products have been designed.
What team-native AI really means
Not long ago I was in a stakeholder meeting where two client stakeholders were presenting strategy drafts for their respective products. Both told the room they’d fed the company-approved chatbot, the everyday tool most of their colleagues use, with documents and inputs to arrive at the first draft. Neither draft was wrong. What struck me was the lack of consistency and contextual awareness of the overall company goals and needs. Two employees, two products, two drafts shaped by the same LLM, but very little shared thinking between them.
This won’t be solved by organisations alone, or by AI vendors alone. Organisations need to build the human infrastructure for collective judgement, shared language, common context, a way of deciding together. But OpenAI, Anthropic, Google and Microsoft need to answer a harder question: what does it mean to design AI for a team, not a person?
Today’s AI platforms are exceptional at learning individuals. It’s what they were built to optimise for. The next design challenge for mass-market LLMs isn’t making AI smarter — it’s designing for contextual collective intelligence, not just individual productivity. Until that happens, every enterprise AI rollout is quietly trading coherence for convenience, and most organisations won’t notice until the bill comes due.
If you’re running AI adoption in your own organisation, the test is simple: pull AI-generated outputs from two teams working toward the same goal and see how far apart they’ve drifted. If you can’t explain the gap, coherence is already leaking.
