Microsoft Wants to Make You Addicted to AI. We've Seen This Movie Before.
- 1 day ago
- 7 min read
Leaked internal documents reveal Microsoft's Scout AI was designed to hook you. The company that rewired social media is now coming for your inbox.
On the morning Microsoft unveiled Scout — its new "always-on personal AI agent" for Microsoft 365 — a different document about the same product was already making its way through newsrooms.
The internal strategy paper, obtained by 404 Media, listed the first phase of Scout's rollout under a single imperative: "Make people addicted."
Not "build engagement." Not "improve retention." Not even the euphemistic "drive daily active users." The word was addiction. Used deliberately, by name, by two senior Microsoft executives — one of whom, Corporate VP Omar Shahine, had published Scout's official launch blog on Microsoft's own website just hours earlier.
The public message: Scout is a human-centered productivity tool designed to give you time back.
The internal message: three phases, from addictive app to agentic platform.
We've heard this before. And we know exactly how it ends.
What the Documents Actually Say
The leaked document, titled "ClawPilot: Overview and Plan with Project Lobster" (ClawPilot being Scout's internal name before launch), describes a three-phase rollout strategy for the product. Phase 1 is not "achieve product-market fit" or "validate core use cases." It is, verbatim: "Make people addicted."
The document goes on to describe the goal of building "the skill and tool ecosystem that makes people depend on it daily." It notes, approvingly, that this addiction "is already happening organically" among the more than 1,000 Microsoft employees currently using the internal version — including, reportedly, CEO Satya Nadella himself.
Phase 2 expands Scout's capabilities. Phase 3 turns it into a broader AI agent platform. But the foundation for all of it, the thing that has to work before any of the rest matters, is dependency.
Microsoft's official response was remarkable in its transparency. Nadella told The Information he was "not sure what this document is or who is writing and leaking this nonsense." 404 Media promptly noted that the document was written by his own executives, one of whom ran Scout's public launch.
The denial didn't stick.
A separate statement to Android Authority tried a softer approach: "Our goal isn't more screen time. It's more time back." That's a nice line. It's also directly contradicted by a document written by the people building the product.
This Is the Social Media Playbook. Exactly.
If you're experiencing a sense of déjà vu, you should be.
In 2021, Frances Haugen — a former Facebook data engineer — leaked thousands of pages of internal documents to Congress and the Wall Street Journal. What those documents showed was that Meta had known for years that its platforms were causing measurable psychological harm, particularly to teenage girls, and had chosen engagement over safety at every fork in the road.
Internal researchers described Instagram as "a drug." One memo noted that the company was "basically pushers." Another showed that Instagram worsened body image issues for one in three teenage girls — and the response wasn't to fix the product, it was to suppress the research.
The features that created this harm weren't accidents. Infinite scroll, autoplay, notification defaults, variable reward loops — these were design choices, made deliberately, to maximize the time users spent on the platform. Not because more time on the platform made users happier. Because more time on the platform made Meta richer.
The result: a teen mental health crisis that researchers are still trying to fully quantify, hundreds of active lawsuits, a $5 billion FTC fine, and ongoing congressional scrutiny. Meta spent years publicly denying that its platforms were designed to be addictive while internal documents showed exactly the opposite.
Microsoft just skipped the denial phase and put it in the document.
The FTC is now investigating AI chatbots and their design practices — specifically whether features that mimic human relationships and encourage compulsive use are being deployed in ways that harm users, particularly minors. The era of "engagement at all costs" is, slowly, ending. Microsoft's Scout documents landed directly into that regulatory environment.
"Addicted" Is Not a Metaphor
When Microsoft's internal document uses the word "addicted," it's worth being precise about what that means in practice.
Scout is not a chatbot you open and close. It's described as an "always-on personal agent" that continuously reads your Teams messages, emails, calendar invites, OneDrive files, and browsing history — building a model of you, your work, your relationships, and your habits that it uses to proactively surface information and complete tasks on your behalf.
The addiction model for a product like this isn't about making you spend hours talking to a chatbot. It's about making you structurally dependent — creating a situation where the product has so thoroughly embedded itself in your workflow that removing it feels disruptive or impossible. The same way people can't stop checking email not because email is pleasurable but because the cost of not checking feels too high.
That's the dependency the document describes. "Build the skill and tool ecosystem that makes people depend on it daily." Not enjoy it. Depend on it.
One anonymous Microsoft employee who spoke to 404 Media called the language "very troubling." They added: "We're seeing more and more addiction happening with AI chatbots and agents overall, and addiction to me is something no product should be making a part of its build strategy."
That employee is right. And the fact that they felt they had to say it anonymously tells you something about the culture that produced this document.
There's a Cost That Doesn't Show Up in the Product Strategy
Here's what Microsoft's three-phase addiction roadmap doesn't mention: every query Scout processes has an energy cost.
Scout is an always-on AI agent that continuously monitors and processes data across Microsoft 365. It's not making one API call when you ask it a question. It's running inference constantly, in the background, on your behalf. That inference runs on Microsoft's Azure data centers, which — despite Microsoft's sustainability commitments — have been expanding faster than their renewable energy capacity can keep up with.
Microsoft quietly delayed its 2030 carbon-negative pledge in 2024 as AI infrastructure buildout accelerated. The company's water consumption for data center cooling has increased sharply. The gap between its public climate commitments and its actual infrastructure trajectory is one of the more underreported stories in the AI industry.
When you design a product to maximize dependency, you're not just optimizing for engagement metrics. You're optimizing for energy consumption. More usage means more queries. More queries mean more compute. More compute means more electricity, more cooling, more carbon.
Designing for addiction isn't just a mental health problem. It's an environmental one. And unlike the mental health consequences — which are diffuse, delayed, and hard to attribute — the energy math is straightforward.
The social media industry spent a decade extracting attention from users. The AI industry is building the infrastructure to do the same thing, at greater scale, with higher energy costs per interaction, and with the added dimension of continuous passive monitoring rather than active use. The environmental footprint of a genuinely addictive AI agent, deployed at Microsoft 365 scale, is not a small number.
The Alternative Isn't No AI. It's Better AI.
None of this is an argument against AI assistants. Used well, they're genuinely useful. The problem isn't the technology. It's the incentive structure that produces documents like Microsoft's.
When the primary KPI for a product is addiction, every design decision gets optimized around that goal. Features that create dependency get shipped. Features that help users accomplish tasks and then get out of the way don't. The product becomes increasingly good at hooking you and decreasingly good at actually serving you — which is, of course, exactly what happened with social media.
The alternative is to build AI products around a different set of incentives. Not engagement. Not dependency. Not screen time. Actual usefulness — the kind that helps you do something, then lets you go live your life.
This is harder to build. It's harder to monetize in the traditional VC-backed growth-at-all-costs model. And it produces different metrics — ones that don't show up in a three-phase addiction roadmap.
At Viro AI, we think about this constantly. Our mission isn't to maximize your time in the app. It's to help you accomplish what you came to do and then fund something that matters with the energy that query consumed. Every Viro query funds renewable energy and climate restoration projects — because we believe that if AI is going to have an environmental cost, that cost should be offset by something real.
We're not trying to make you addicted. We're trying to make you more effective, more informed, and less guilty about using AI in the first place.
Those are different products. Built on different values. For different reasons.
One of them just put "make people addicted" in a strategy document. The other is trying to plant trees with the electricity bill.
The Quiet Part Out Loud
What's most striking about the Microsoft Scout leak isn't that the company wanted to maximize engagement with its product. Every tech company wants that.
What's striking is that someone wrote it down. With those words. In a document attributed to the executive who also wrote the public launch announcement.
There's a version of this where it's just sloppy internal language — a phrase that was meant to convey "build a habit-forming product" and got expressed too bluntly. Microsoft's defenders have made this argument. Maybe.
But "make people addicted" is a specific phrase with a specific meaning. It's the phrase the social media industry spent years denying it had ever used, even as internal documents showed exactly the opposite. It's the phrase that is now at the center of thousands of lawsuits and years of congressional testimony.
Microsoft didn't use it by accident. They used it because that's what they were trying to do.
The difference this time is that we have the document on day one. We don't have to wait for a whistleblower five years down the line. We don't have to wait for the mental health crisis to manifest before we understand the design philosophy behind it.
We know now. The question is what we do with that.
Viro AI is the anti-Big Tech AI assistant. We fund renewable energy and climate restoration with every query — because we think AI should help the world, not hook you to a screen.
Comments