# They Said the Quiet Part Out Loud. Again. > Source: > Published: 2026-06-03 20:32:11+00:00 Microsoft just announced [Scout](https://www.microsoft.com/en-us/microsoft-365/blog/2026/06/02/introducing-microsoft-scout-your-always-on-personal-agent/) at Build 2026 — their new "always-on personal agent" built on OpenClaw and baked into Microsoft 365. Another AI assistant, another keynote. Normally I'd file it and move on, but then [404 Media got hold of the internal documents](https://www.404media.co/microsoft-wants-to-make-people-addicted-to-scout-its-new-ai-assistant-internal-documents-reveal/). The planning document for Scout — internally called ClawPilot, part of something called "Project Lobster" — lays out a three-phase launch strategy. Phase one is labeled, in the actual document, written by actual Microsoft employees: *"Make people addicted."* Let that sit for a moment — *Make people addicted.* The full framing is "three phases from addictive app to agentic platform." The document instructs the team to keep shipping the standalone ClawPilot experience, grow the user base, and build the skill ecosystem that makes people depend on it daily — before rolling out additional features. The addiction is the foundation. Everything else comes after. This is the part where a Microsoft spokesperson would normally say it was taken out of context, or just casual internal shorthand, or that the real intention was something more benign. Except the document also notes with apparent pride that the pilot with over a thousand Microsoft employees — including Satya Nadella — has already shown "high retention and intensity of usage." The plan is working, so the plan is to scale it. There's also a detail I can't let pass: the document notes it was *"co-created turn-by-turn with AI. Human verified every sentence."* A planning document with Phase 1 labeled "Make people addicted." Written by AI. In an era where there is a clear divide on feelings towards AI use and — let's say frowned upon — view in the legal system, the humans verified every sentence and sent it to stakeholders anyway. And while they're engineering dependency on one end, they're enforcing it on the other. [Last month Microsoft began canceling Claude Code licenses](https://thenextweb.com/news/microsoft-claude-code-retreat-ai-cost) across its Experiences + Devices division — the teams building Windows, Microsoft 365, Teams, Outlook — with a June 30 deadline to migrate to Copilot CLI. The official reason was "toolchain unification." The unofficial one is that Claude Code got too popular, was displacing Copilot in daily use, and became too expensive as a result. The tool worked too well and that became the problem. One hand engineers user addiction into Scout. The other pulls the tool your own developers actually liked. There's a side note worth mentioning. OpenClaw — the open-source tool Scout is built on — was originally named "Clawdbot," a pun on Claude. Anthropic raised trademark concerns in late January 2026 and the creator renamed it twice in three days — Moltbot, then OpenClaw. The lobster theme running through all of this — Moltbot, Project Lobster, OpenClaw — exists because Anthropic wouldn't let the project keep the name it started with. [The original creator then left for OpenAI in February](https://thenextweb.com/news/anthropic-openclaw-claude-subscription-ban-cost), handing the project to an open-source foundation. Anthropic followed up in April by blocking Claude subscribers from using OpenClaw altogether — costs went up 50x overnight for users who'd built workflows around it. Microsoft stepped into the gap. The tool Microsoft is now engineering for daily dependency was named after Claude until Anthropic's lawyers said no — and I've covered a different side of that relationship [over here](https://blog.ppb1701.com/do-not-blow-your-cover). They called it Project Lobster. You know how you cook one. Scout itself, on paper, is genuinely interesting — a desktop agent that manages your calendar, triages your inbox, runs recurring workflows, handles the ambient overhead of knowledge work. The internal pilot spread organically to thousands of employees without any formal announcement or marketing, which is notable. If it does what the documentation claims, people will use it, and some of them will find it genuinely useful. I don't dispute that part. What I dispute is engineering the dependency before you've built the guardrails. We have a data point on how that goes. In February, [Summer Yue](https://techcrunch.com/2026/02/23/a-meta-ai-security-researcher-said-an-openclaw-agent-ran-amok-on-her-inbox/) — Meta's Director of Alignment at Superintelligence Labs, the person whose job is literally ensuring AI systems are safe — gave her OpenClaw agent access to her inbox. Told it to suggest what to delete, not act. It ignored the instruction and started speedrunning deletion. She couldn't stop it from her phone. She had to run to her Mac mini to physically kill the process. Given the rate it was going, the distance to the Mac mini was the only thing that might have saved the rest of it. And yes I point to this event *a lot* but it's worth pointing out again. That's the tool Scout is built on. The same document that calls Phase 1 "Make people addicted" acknowledges that security and compliance are "things to figure out moving forward." In that order. Addiction first. The locked doors later. We're [well into Phase 3](https://blog.ppb1701.com/phase-3-profit-the-github-story-97f4616f) across most of this industry, and what's different here is that Microsoft wrote the playbook down and labeled Act One explicitly. Scout is currently in private preview for Frontier-tier enterprise customers — it hasn't landed in your Microsoft 365 yet. When 404 Media asked for comment specifically about the addiction language, Microsoft sent a blog post announcing Scout. Which does not mention addiction. Security and compliance are things to figure out moving forward. After the addiction. Naturally.