Microsoft Adopted OpenClaw, Google Cloned Its Design, Meta Wants to Sell a Consumer Version: Three Strategies for the Same Agent

A year ago, none of the world’s largest tech companies would touch OpenClaw. Microsoft’s CEO called shipping it the equivalent of “launching a virus.” Security researchers flagged plaintext API key leaks, a malware-riddled skills marketplace, and a prompt injection attack that wiped a Meta executive’s entire email inbox. Now Microsoft runs on OpenClaw’s code, Google has rebuilt its blueprint, and Meta is developing a consumer clone.

Ken Yeung’s analysis in The AI Economy traces how three companies with different structural incentives arrived at three different answers to the same question: what do you do when an open-source project defines the category you need to own?

Microsoft: Adopt the Code, Build the Cage

Microsoft’s strategy is the most direct. At Build 2026, the company announced OpenClaw would power Scout, its first “Autopilot” agent for Microsoft 365, and run natively on Windows inside Microsoft Execution Containers (MXC). MXC provides process-level isolation with policy-driven containment enforced at the operating system level.

Satya Nadella framed the containment work as what made adoption possible. “It’s so wonderful to see OpenClaw come to Windows and have all that capability in terms of the security and that comfort to be able to have these long-running Agents and unmetered intelligence come together,” he said during his Build keynote, as Yeung reported.

The competitive logic: developers were already running OpenClaw on Windows without Microsoft’s blessing. Rather than fight that adoption, Microsoft chose to make its platform the safest place to run the agent. It is also contributing its policy conformance system back to the open-source project, giving it influence over what “secure OpenClaw deployment” means for every organization running it.

Microsoft gets distribution (Scout boosts M365 subscriptions), governance revenue (enterprise security tooling), and upstream influence, all without owning the project.

Google: Clone the Design, Kill the Open Source

Google took the opposite path. In March, it released a way for developers to integrate OpenClaw agents into Google Workspace, acknowledging that users were bringing their own agents whether Google built for them or not. Then at I/O in May, it introduced Gemini Spark, an always-on agent that composes emails, updates study guides, and monitors credit card statements, powered by Gemini 3.5 with Google’s Antigravity harness underneath. The capabilities mirror OpenClaw’s. The code does not.

The same month, Google announced that its open-source Gemini CLI, an Apache-licensed tool with hundreds of contributors, would stop working with Google AI subscription plans on June 18, replaced by the closed-source Antigravity CLI.

As one developer observed, the result is a vision where “every agentic workflow you build runs on infrastructure Google controls, through interfaces Google designs, against models Google hosts.”

Google’s structural incentives explain the approach. It has spent years building the full stack: its own chips, models, and agent harness. Constructing a proprietary OpenClaw equivalent was the natural move. Whoever controls the agent sitting in front of Gmail, Docs, and Calendar controls the user relationship, and Google has no intention of ceding that to someone else’s software.

Meta: Build a Consumer Clone, Charge $200 per Month

The Information reported that Meta is developing Hatch, a consumer-facing OpenClaw alternative that could cost up to $200 per month. Unlike Microsoft’s enterprise-focused Scout, Hatch targets non-technical users, with Meta testing it on simulated versions of DoorDash, Reddit, Etsy, and Outlook.

Meta’s relationship with OpenClaw’s creator adds context. Mark Zuckerberg personally offered feedback to Peter Steinberger and tried to hire him before Steinberger joined OpenAI in February, according to Business Insider.

Whether Meta will fork OpenClaw or build Hatch entirely on its own models remains unclear. A launch could come as early as next month.

Three Bets on Agent Control

The split between these three companies reflects a deeper disagreement about where value accrues in agent infrastructure.

Microsoft is betting that the agent runtime becomes a commodity and value concentrates in the governance, security, and enterprise integration layers. Google is betting that vertical integration, from silicon to model to agent to application, creates a moat no open-source project can replicate. Meta is betting that consumer willingness to pay $200 per month for a personal agent justifies building a closed alternative to a free one.

All three strategies validate the same premise: OpenClaw defined what a personal AI agent looks like, and every major platform now needs an answer to it. The question for builders is which cage they’re willing to work inside, and whether the open-source version, now stewarded by the OpenClaw Foundation Steinberger created before leaving, can remain independent when every major tech company has a reason to co-opt it.