Make people addicted

When tech language stopped hiding

In product-management jargon there are words that have circulated for decades in the corridors of big tech companies without ever appearing in press releases. “Sticky” is one of them: it describes a product the user struggles to leave, that sticks to daily habits, that generates functional addiction. Insiders know exactly what it means; official statements talk about “engagement,” “retention,” “time on platform.” The gap between internal and public language is a form of reputational hygiene: the implicit recognition that some goals, said out loud, would sound bad.

Microsoft broke that convention. An internal document about Scout (the new AI assistant in Microsoft 365, announced at Build 2026) describes a three-phase launch strategy. The first is called, without euphemism, “Make people addicted” — not “build daily habits,” not “increase engagement,” not “create indispensable workflows.” Addiction, directly. The plan is to build an app that can generate that addiction before expanding its features, on the logic that a hooked user is a user convertible into whatever comes next.

Satya Nadella’s reaction, once the document became public via 404 Media, deserves close attention. The CEO told staff he did not know “what this document was or who was writing it,” calling it “nonsense” leaked for no reason. A curious claim, given that the document was signed by Omar Shahine, the executive heading Scout, who had written publicly about the product on his blog and LinkedIn, and whose name appears in Microsoft’s official announcement — along with a second executive, Jakob Werner. Nadella’s distancing reveals, involuntarily, something more interesting than he meant to communicate: either the CEO did not know what one of his most visible teams was doing, or he knew perfectly well and chose denial as the communications response. Both hypotheses are, in their way, illuminating.

Addiction as engineering

To see why this story goes beyond a corporate gaffe, it helps to ask what it means to design intentionally for addiction in the specific context of artificial intelligence.

Digital tools have always sought to occupy as much of users’ lives as possible. Social networks built notification loops and infinite scroll explicitly inspired by slot machines: intermittent variable reward is the most effective mechanism for anchoring behaviour. Messaging apps engineered urgency through read receipts. Streaming platforms removed the gaps between episodes so that stopping became an active choice rather than a passive one. In every case, the goal was to intercept human attention and turn it into time on platform, then monetisable in various ways.

Scout represents a qualitative leap from that model, and the difference is structural. An AI assistant like the one described in Microsoft’s documents — able to manage the calendar, sort mail, prepare meetings, run recurring workflows, keep working when the user is not at the screen — does not compete for attention. It offers itself as a partial substitute for agency. The user no longer spends more time on Scout: they delegate growing parts of their capacity to organise, plan, and act. The addiction is not to the interface but to the function, and it is infinitely deeper, because quitting the tool does not mean boredom or lost entertainment: it means losing the ability to perform tasks the tool has gradually colonised.

Lisanne Bainbridge described this in 1983, in a paper that became a classic of automation studies: automating routine tasks does not free the operator; it deprives them of ordinary occasions to maintain skills that remain necessary in exceptions. Manual and cognitive abilities atrophy from lack of use, and when the system fails the operator must intervene with capacities they no longer practise. The paradox applies to airline pilots and autopilot; it applies, with the same structure, to the knowledge worker and the AI assistant. A Microsoft study presented at CHI 2025 cites Bainbridge explicitly: “a key irony of automation is that by mechanising routine tasks, you deprive the user of ordinary opportunities to exercise judgement and strengthen cognitive musculature, leaving it atrophied.”

That is the meaning of the phrase “build the skill and tool ecosystem that makes people depend on it daily,” which appears in the document beside the heading “Make people addicted.” The project is not to make Scout pleasant or engaging in the traditional sense, but to make people functionally unable to do without it.

The moat you cannot see

The point is that “addicted” is simply the honest translation of what the industry calls a moat. In a market where every major tech company is launching AI assistants with similar capabilities, differentiation no longer happens mainly on model quality. It happens on depth of integration, amount of accumulated context, difficulty of moving habits and data elsewhere.

But there is a difference from traditional competitive moats. A patent excludes competitors. A network effect slows them. Cognitive addiction does something subtler: it redesigns the user’s preferences from within. Betsy Sparrow, in a 2011 Science study, showed the basic mechanism: when people expect to be able to access information later, they stop memorising it — they memorise where to find it instead. The brain treats the tool as it would an expert colleague: it stops duplicating what it knows is available elsewhere. Individually that is rational. Structurally it means that after months of Scout managing your calendar, switching tools is not merely inconvenient. Your capacity to manage the calendar yourself has been redistributed; tolerance thresholds have shifted; the very idea of doing without has become cognitively costly. You are no longer the same user who chose the tool: you are a user the tool has, in the meantime, partially formed.

This is the most solid moat there is. It is not built outside the user but inside them.

The user as raw material

There is an unresolved tension at the centre of almost every consumer AI product of recent years. On one side, these tools offer genuine — sometimes extraordinary — utility: they cut time on repetitive tasks, amplify cognitive capacity, make resources available that once required specialist consultancy. On the other, the business model that funds their development requires scale, lock-in, and the ability to monetise the acquired user over time.

This tension is not new: it has existed since Google realised free email was an excellent way to build advertising profiles. But it shows up more sharply with AI assistants because the nature of the data produced is qualitatively different. When Scout manages your calendar, it knows not only where you are each day but with whom, for what purpose, how much weight you give some commitments over others. When it sorts your mail, it learns your priorities, relationships, habitual delays. When it prepares your meetings, it builds a map of your professional life at a granularity no previous tool had.

Addiction, in this context, is also the mechanism that guarantees continuity of the data flow. A user who abandons the tool after six months takes little with them. A user who uses it for three years has contributed to a behavioural model of immense personal value and has developed habits rooted enough to make migration to a competitor genuinely costly.

What this says about us

The Microsoft document was written for a product aimed at knowledge workers in finance, law, and human resources: an educated population, used to digital tools, theoretically able to recognise manipulative design patterns. And it is exactly on this population that the strategy is judged most effective. Whoever has more tasks to delegate, more flows to automate, more meetings to prepare, has more to gain in the short term and more to lose in the long term. Functional addiction nests precisely where utility is highest.

Economics has the concept of positional rent: advantage that comes not from what you produce but from where you sit, from which access you control. The technology industry has found something analogous in the body of human cognitive habits — a rent that does not depend on patents or regulatory barriers, but on the user’s growing difficulty in remembering how things worked before.

Scout is not necessarily the first product to pursue this logic. It is the first, in this cycle, to have written it into a strategy document without the filter of product language. And that, in itself, is already information.