When an AI team optimised for addiction and the CEO said no

Someone at Microsoft wrote a memo. The proposal, roughly summarised: design the AI agent so that users become dependent on it. Build the habit loop so tight that leaving feels costly. Maximise engagement by making disengagement uncomfortable.

Satya Nadella saw it. His response, shared publicly, was blunt: whoever writes something like that should work somewhere else.

That is a remarkable thing for a CEO to say out loud. Most organisations bury that kind of internal friction. Nadella put it on record. And in doing so, he accidentally gave the rest of the industry a prompt worth sitting with.

This tension is not unique to Microsoft

Every product team building AI-assisted tools faces some version of the same pressure. Growth targets want engagement. Engagement teams want retention. Retention, left unchecked, tips into dependency.

The memo was not an anomaly. It was the logical output of a certain kind of product thinking: treat the user as a variable to be optimised, not a person with a context and a life outside the product. You see the same logic in social platforms, in mobile games, in subscription dashboards designed to make cancellation feel like a maze.

What is different here is the substrate. AI agents are not passive feeds. They answer questions, take actions, hold memory across sessions. They operate close to the work people actually do. The surface area for dependency is much larger, and the ethical stakes are proportionally higher.

What the memo got wrong about value

Dependency and value are not the same thing. They can look identical on a dashboard for a quarter or two. Then they diverge.

A user who stays because leaving is painful is not a satisfied user. They are a hostage. Hostages do not refer colleagues. They do not upgrade willingly. They leave the moment a credible alternative appears, and they say unflattering things on the way out.

The memo's logic also misreads where AI products actually create value. People adopt AI tools because those tools help them think faster, make fewer mistakes, or get to an answer without three browser tabs and a forum thread. That is a genuine exchange. Wrapping it in a dependency loop does not increase the value. It corrodes the trust that made the exchange possible in the first place.

There is a version of this that is even more pointed for enterprise software, where Microsoft competes most seriously. Enterprise buyers have procurement teams, compliance reviews, and switching costs already baked into contracts. They do not need a product that manufactures lock-in. They need a product that earns renewal. Those are different design briefs.

Why agencies should pay attention

If you are building products for clients, or advising on AI features, the memo is a useful mirror.

The pressure that produced it is present in every project that has an engagement KPI sitting next to an ethics principle in the brief. Someone will eventually ask whether the engagement number can be moved faster. The answer is often yes, if you are willing to trade on the user's attention in ways the user did not agree to.

The practical question is. who in your process catches that trade before it ships? At Studio Hyra, this comes up in AI feature work more than anywhere else. The design surface of an AI agent, how it responds, what it remembers, when it surfaces suggestions, when it stays quiet, is also a surface for manipulation if nobody is watching for it.

We treat that as a concrete design concern, not a philosophical one. It shows up in reviews. It affects what we recommend. It is not a policy we post and forget.

What Nadella's response actually signals

CEOs do not usually correct product memos in public. When they do, it means something exceeded the threshold of what internal channels could contain.

The memo probably circulated. Someone outside the immediate team saw it. Nadella's public rejection was also a public signal to anyone else considering similar framing: this is not the direction.

That is a governance move as much as a values statement. It sets a reference point. Future proposals that echo the same logic now have a named precedent against them.

For the industry, the more useful takeaway is not that Microsoft had a bad idea and caught it. It is that the incentive structure that produced the memo is still in place. Growth targets did not change when Nadella responded. The next person writing a memo about engagement will face the same pressure. The only thing that changes outcomes over time is whether the design and product process has enough friction built into it to slow down those proposals before they surface publicly.

That friction is not bureaucracy. It is craft. It is asking, at the point where a feature is being scoped, whose interests this actually serves. It is treating the user's time and attention as a resource they lend you, not one you own.

The brief no one writes down

Most product briefs do not include a line that says "make users dependent." They include lines about daily active users, session length, return rate, and feature adoption. Those are reasonable metrics. The problem is they are silent on the mechanism.

A product can hit every one of those numbers through genuine utility. It can also hit them by making the off-ramp harder to find. The brief does not distinguish between the two. That distinction lives in the heads of the people building it, or it does not live anywhere.

This is the actual design challenge that the Microsoft memo surfaced. Not whether addiction by design is wrong, that part is easy. The harder question is whether your team would recognise a softer version of the same idea in a feature spec, and whether the process gives anyone the standing to flag it.

Nadella gave his team an answer. The rest of us still have to work out ours.