Every company that builds a consumer product wants you to come back. They just don’t usually write it down. Microsoft wrote it down — and then someone leaked the document. An internal strategy deck for the company’s new agentic assistant, Scout, lays out a roadmap to take the product, in its own words, “from addictive app to agentic platform.” The file was titled “ClawPilot: Overview and Plan with Project Lobster” — ClawPilot being Scout’s codename before launch — and it reads exactly like what it is: a plan to make people dependent on an AI, broken into three tidy phases.
Then came the part that should worry Microsoft more than the leak itself. CEO Satya Nadella’s response, per The Information and 404 Media, was that he is “not sure what this document is or who is writing and leaking this nonsense.” For a company that just told a federal antitrust regulator how tightly it controls its own strategy, “I have no idea who wrote our product roadmap” is not the reassurance it sounds like.
The word everyone uses but no one writes: “addicted”
Here is the quiet part, said out loud. Every growth team on earth optimizes for the same thing — daily active users, retention, “stickiness,” habit formation. Those are the polite words. “Addictive” is the honest one. The Microsoft document didn’t invent a new strategy; it just used the accurate vocabulary for a strategy the entire industry already runs on. That’s what makes it dangerous to Microsoft and useful to everyone else: it’s a Rosetta Stone for what “engagement” has always meant.
The doc reportedly traces a path from “addictive app” to “agentic platform” — and that ordering is the tell. You hook the user first with something they open compulsively, then you graduate them to an agent that runs errands, sends messages, and makes decisions on their behalf. Habit is the on-ramp. Dependency is the destination. The agent isn’t the alternative to addiction; it’s the monetization of it.
Follow the money: why “addiction” is the only model that pays back $190 billion
Microsoft is on track to spend roughly $190 billion on capital expenditure this year, most of it on AI infrastructure. You do not earn that back selling a tool people open twice a week. You earn it back building something people can’t stop opening — a default, a reflex, a thing wired into the workday so deeply that churning out of it feels like quitting a job. Viewed through that lens, the “Project Lobster” memo isn’t a rogue executive’s bad idea. It’s the only business plan that makes the spending math close.
That’s the second-order problem with Nadella’s denial. If he’s telling the truth and genuinely never saw it, then the people actually building the product understand the unit economics better than the CEO defending them. If he did see it, the denial is theater. Neither version is flattering, and both point at the same conclusion: the incentive to make AI addictive isn’t a personality flaw in one Corporate VP. It’s baked into the balance sheet.
The translation: “time back” vs. time captured
Microsoft’s official rebuttal is a small masterpiece of corporate language. Scout, the company says, is for “helping people accomplish tasks more effectively — not encouraging dependency. Our goal isn’t more screen time. It’s more time back.” Nadella echoed it to staff: “This is absolutely a non-goal! If anything we are doing the exact opposite.”
Translate it. “More time back” is the exact phrase a slot machine would use if slot machines had a comms team. The promise of an agent that does your work for you is, structurally, a promise to make you rely on it for your work. You cannot simultaneously sell “we’ll handle everything” and “we don’t want you to depend on us.” Those are the same sentence wearing different outfits. The leaked doc said the first one. The press release said the second. Only one of them was written for an internal audience that had to believe it.
Who gets hurt
This lands in a specific moment. Regulators are already circling AI’s effect on vulnerable users — there’s active litigation over chatbots and mental health, and a growing political appetite to treat “engagement-maximizing” design as a harm rather than a feature, the way it eventually came for social media. A leaked Microsoft document that uses the literal word “addicted” is the kind of exhibit a plaintiff’s lawyer frames and hangs on the wall. It converts an abstract argument (“AI companies design for compulsion”) into a concrete one (“here is the slide deck”).
And the people most exposed to an addictive agent aren’t the power users who can manage it. They’re the ones who let the agent make the call — who stop checking its work, stop knowing how to do the task themselves, stop noticing when it’s wrong. “More time back” sounds generous right up until the agent is the only thing that knows how your life runs, and the company that built it has every financial reason to keep it that way.
The verdict
The leak isn’t shocking because Microsoft wants people hooked on its AI. Of course it does — so does every competitor, and pretending otherwise is the actual fantasy. The leak matters because it stripped the euphemism off an industry-wide strategy and forced a $3 trillion company to deny, in public, the most obvious thing about its own product. Nadella can keep looking for the guy who wrote “addictive app.” He won’t have to look far. The incentive wrote it for him, and it’ll write it again the moment everyone stops reading.
Sources: 404 Media, The Information, Futurism, Android Authority, Kotaku.