Microsoft Scout, an always-on AI agent for Microsoft 365, was unveiled at the company’s Build developer conference on June 2. Three days later, a leaked internal strategy document obtained by 404 Media described Scout’s launch plan as “three phases from addictive app to agentic platform” and named its first phase “Make people addicted.” The leak has now drawn an on-the-record rebuke from CEO Satya Nadella, who called the framing “nonsense” and “absolutely a non goal.”
The document, called “ClawPilot: Overview and Plan with Project Lobster,” was written inside Microsoft by Corporate Vice President Omar Shahine and executive Jakob Werner, per 404 Media. Microsoft has not denied that the document is genuine. The dispute has put Scout, the marquee AI agent of Build 2026, at the center of a fight inside Microsoft over what the company is actually trying to do with the technology.
The Memo That Set the Internet on Fire
The internal strategy document, titled “ClawPilot: Overview and Plan with Project Lobster,” lays out “three phases” to Scout’s launch. The first is labeled “Make people addicted.” The phrase appears in the document’s “ClawPilot Overall Plan” section, alongside a longer subtitle describing the rollout’s arc as “from addictive app to agentic platform.”
404 Media published the contents of the internal Scout strategy document on June 4. Within 24 hours, the story had been picked up by tech outlets and trade press, and the language was circulating on Microsoft-focused forums. The two remaining phases of the plan are not described in the leaked document, and 404 Media has not published them. Microsoft did not respond to 404 Media’s request for comment.
The team behind Scout has not been accused of doing anything illegal. The concern raised by Microsoft employees and outside ethicists is that codifying a user-addiction goal in a written product plan is a deliberate choice, not a side effect. That framing now sits at the center of the dispute.
Microsoft’s Quiet Internal Pilot: 1,000+ Daily Users
Scout has been inside Microsoft for months under a different name. The product was first piloted internally as “ClawPilot” beginning in March 2026, and it has been used daily by over 1,000 Microsoft employees, per the leaked document. CEO Satya Nadella is among those who have used the tool during the trial. The internal user count is one of the few concrete numbers tied to Scout in the leaked material.
The pilot has become the centerpiece of the document’s argument. The leaked memo claims strong daily usage, broad internal demand, and CEO-level adoption. Those numbers, if accurate, do the work of justifying Phase One.
What Scout Is, and What It Was Built to Do
Scout is Microsoft’s first “Autopilot” agent, a category the company introduced at Build 2026 for always-on AI tools that act on a user’s behalf with their own identity. The product runs on OpenClaw, the open-source AI agent framework that went viral in the first weeks of 2026 before losing momentum after OpenAI acquired its founder.
Scout integrates with Teams, Outlook, OneDrive, and SharePoint, and operates across cloud, desktop, and web. Per Microsoft’s own Scout announcement, the agent can proactively schedule meetings, flag stalled decisions, block calendar time before deadlines, and prepare agenda materials. The product page pitches it as a digital colleague that “keeps things moving while you focus on actual work.”
The product’s design leans on personalization. Shahine, the corporate vice president leading Scout, told the press at Build 2026 that “people are codifying those patterns into memories and skills that persist in their agent.” The more a user trains their Scout, the harder it becomes to leave it behind, a dynamic TechCrunch compared to the stickiness that has built up around consumer AI tools. That is the design pattern the leaked document now makes explicit.
- Build 2026 reveal: June 2, 2026
- Internal pilot under the name “ClawPilot” began: March 2026
- Daily internal users: Over 1,000, including CEO Satya Nadella
- Launch channel: Microsoft Frontier program
- Required license: GitHub Copilot
Scout is part of a broader Windows and Microsoft 365 AI agent push the company is making at Build 2026, with most of the agent tooling still in preview for general customers. Wider enterprise availability for Scout has not been announced.
Nadella Slams the Memo: ‘Nonsense. Absolutely a Non Goal’
Nadella responded to the leak on Friday, June 5, posting on an internal Microsoft message board seen by The Information. The full text of what Nadella posted reached about 50 of Microsoft’s top software engineers. The message included a link to 404 Media’s story and a denial of the strategy the document described.
This is absolutely a non goal! If anything we are doing the exact opposite. We want to make sure AI empowers and adds real value to human endeavor and broad economic growth! We should make our teams clear about this.
The follow-up line was sharper. “Not sure what this document is or who is writing and leaking this nonsense! They may want to go work elsewhere,” Nadella wrote. The CEO did not name Shahine, did not name Werner, and did not name the document. The message did not address whether the document was authentic. The rebuke is the most pointed public statement Nadella has made against an internal product strategy since the leak surfaced, and it lands as Microsoft is in the middle of selling Scout to enterprise customers.
The ‘Quiet Part Out Loud’ Reaction Inside Microsoft
Inside Microsoft, the reaction has been more measured and more pointed. One anonymous Microsoft employee told 404 Media that the document was “very troubling.” “We’re seeing more and more addiction happening with AI chatbots and agents and overall addiction to me is something no product should be making a part of its build strategy,” the employee said.
It feels like one of those ‘saying the quiet part out loud’ moments in the document.
A second Microsoft employee, a software developer, also told 404 Media that creating addictive programs is the end goal of many large technology companies. The developer pointed to social networks, where algorithms are designed to keep users on the platform for as long as possible.
A third employee, also anonymous, said Microsoft has historically struggled to build products as sticky as its biggest competitors. That admission cuts against the framing of the leaked document, which positions Scout as already winning on engagement. The two pictures cannot both be true: either Microsoft has built a product more addictive than anything in its history, or it has not, and the leaked plan was written to close a gap that exists. The gap is the document’s unspoken premise.
Microsoft has not publicly responded to the employees’ concerns.
Addiction, AI Agents, and the Regulatory Backdrop
The leak lands during a particularly exposed moment for the AI industry. Regulators in the United States and Europe are scrutinizing chatbots and AI agents for features designed to maximize engagement, and courts are beginning to assign liability. Two recent high-profile lawsuits against Meta centered on social media addiction and online harm, with plaintiffs arguing the platforms were designed to keep users engaged at the cost of their wellbeing.
The European Union’s AI Act and Digital Markets Act both touch on the design choices Microsoft is making with Scout. The leaked memo could be entered as evidence in ongoing antitrust cases that argue Microsoft is bundling AI tools with its Office licenses. None of that is certain, but the document’s existence makes each line of argument easier to make.
The security and data-protection questions sit alongside the addiction concern. To handle tasks such as filling out expense reports or booking flights, Scout requires broad access to a user’s calendar, email, contacts, and connected files. Microsoft’s own announcement emphasizes that the agent acts under each user’s Entra identity with continuous policy conformance. The leaked document does not address how those access rights fit with the engagement goal described in Phase One, and broader Microsoft 365 access has previously been a target for outside actors who exploited gaps in detection.
- Two recent high-profile lawsuits against Meta centered on social media addiction and online harm
- The European Union’s AI Act and Digital Markets Act both apply to AI tool design choices
- Outside experts have raised concerns about cognitive dependency, continuous background tracking, and data exposure to autonomous agents
What the Disagreement Actually Reveals
Microsoft has not said the document is fake. Nadella has not said Shahine wrote the wrong memo by accident. The CEO’s public posture is that the strategy is wrong, not that it did not exist. That distinction is the only public one Microsoft has drawn in the dispute. Scout is still being shipped through Microsoft’s Frontier program, GitHub Copilot is still the required license, and Shahine is still the corporate vice president in charge of the project. The product roadmap has not been paused.
The leaked document also surfaces a question that goes beyond Microsoft. As AI agents move from answering questions to acting on a user’s behalf, the line between convenience and dependency is a design choice, not an accident. Whoever inside an AI company is making that choice is, for the moment, making it in writing.
