How Microsoft is bringing OpenClaw to the masses

Hello again, and welcome back to Fast Company’ s Plugged In . Like most tech-company keynotes, the one this week at Microsoft’s Build developer conference in San Francisco featured a few special guests. They ranged from Nvidia CEO Jensen Huang (who’s in a dead heat with OpenAI’s Sam Altman for the distinction of being the industry’s most omnipresent executive) to music/investing duo The Chainsmokers to Mayo Clinic CEO Gianrico Farrugia. But the most intriguing guest star of the bunch was easily Peter Steinberger , creator of the open-source AI agent platform OpenClaw. Steinberger—whose day job , as of February, is at OpenAI—had good reason to take Microsoft’s stage. Among the two-hour-and-22-minute Build keynote’s major themes was the company’s enthusiasm for OpenClaw and desire to make it accessible to a broader audience. The news reflecting that included a new OpenClaw app for Windows and a technology for sandboxing OpenClaw agents—or “claws”—so they can’t wreak accidental havoc with data . “Watching a claw trying to delete all of your desktop files and just fail makes me really happy,” beamed Steinberger. “Because six months ago, that totally would have worked.” And then there are Autopilots, a new type of Microsoft agent. I didn’t catch their OpenClaw connection at first: Microsoft CEO Satya Nadella mentioned it only briefly and elliptically during the keynote address. Though Autopilots incorporate OpenClaw code and functionality, they’re meant to be safe enough for businesses to roll out without fretting about them destroying or leaking proprietary information. They also operate partially in the cloud rather than assuming you’re willing to run them on a computer that’s powered up 24/7. (For many OpenClaw enthusiasts, that machine is a Mac Mini dedicated to the task .) Microsoft says it plans to release many Autopilots. It’s starting with just one, called Scout , that it’s making available as an experimental release for customers who have opted into its Frontier program . The agent’s purpose, says corporate VP Omar Shahine, “is to give people an assistant that can help them take some work off their plate and manage some of the logistical work that a lot of knowledge workers do—organize meetings, stay on top of things you’ve committed to, [get] helpful reminders.” Along with connecting to Microsoft products such as Outlook, Teams, and OneDrive, Scout can operate a web browser to perform mundane but necessary work tasks such as filing expenses and making travel arrangements. Microsoft Scout, the company’s first OpenClaw-based Autopilot. [Image: Courtesy of Microsoft] Shahine has been at Microsoft for more than a quarter century, but he’s only a few weeks into his position overseeing Autopilots. He readily acknowledges that some Microsoft customers, unnerved by reports of OpenClaw agents running amok , regard the platform as a creator of problems rather than a solver of them. “One of the reasons I’m in this job is because I had a different experience,” he says. “I went through a very intensive experimentation phase where I found a lot of success, and I shared that story with a lot of people internally. I think people’s perceptions changed about the capability and potential.” Despite OpenClaw’s risks, Shahine says that many of the customers he’s talked to have been playing with the technology themselves. Eager to adopt it for real tasks, “they just want someone like Microsoft to take away a lot of the operational concerns that they have.” (On Tuesday, 404 Media’ s Jason Koebler reported on an internal Microsoft document citing “make people addicted” as a goal for Scout. Two days later, The Information’ s Aaron Holmes followed with a story saying the document had been Shahine’s and had provoked an irate response from Nadella. In a statement, Microsoft said the goal was not “more screen time” but “more time back,” which does seem like the only outcome that would make the app useful.) Given that agents are the single buzziest aspect of AI, it would be weird if Microsoft weren’t working on something like Autopilots. Or on several somethings: In March, the company announced Copilot Cowork , another flavor of workplace agent that clearly overlaps with Scout’s purview. But it wasn’t a given that it would base Scout and other Autopilots on OpenClaw technology, rather than seeing the platform as a potential threat to its commanding position in workplace productivity . True, it’s been a long time since the company’s instinctive response to every new trendsetting product was to launch its own me-too version—a Zune for every iPod. In the Nadella era , the company has picked its battles more thoughtfully. Once notorious for using its might to smother open standards—a strategy that became known as “ embrace, extend, and extinguish ”—the tech giant has also grown more comfortable participating in ecosystems it doesn’t control. Shahine calls out one of the best such examples: Microsoft’s 2018 adoption of Chromium, the open-source version of Google’s Chrome, as the rendering engine for its Edge browser . “We had our own web browser for a long time, and I think you know how it turned out,” he says. “It was not good for the broader tech community to have to offer web pages that rendered for Internet Explorer and Chrome and Safari and Opera. When the world sort of standardized on two major rendering engines, things got a lot better for everybody.” That’s not to say that this precedent maps perfectly to what Microsoft is doing with OpenClaw. Adopting Chromium may have been laudable, but it was also a last resort: The company did it only after more than 20 years of building its own browsers from the ground up, a tumultuous era that saw it crush Netscape Navigator and then go on to get walloped by Chrome itself. By contrast, agents are still in their infancy. Steinberger launched OpenClaw only last November, and it didn’t capture the industry’s imagination until January. Microsoft is hardly the only tech behemoth that was quick to see an opportunity in OpenClaw. Shahine notes that companies such as OpenAI, Red Hat, and Nvidia (which has its own tech stack, NemoClaw , designed to make OpenClaw safer) have already chipped in to help propel the project forward. Among Microsoft’s specific priorities, he says, is to make OpenClaw’s code base more manageable, so it’s easier to add new features without breaking existing ones. It’s contributing its work on that front back to the open-source project so that everyone benefits. Right now, the number of people who are actually using OpenClaw is likely microscopic compared with those who are aware of its existence but haven’t taken the plunge. The various OpenClaw-related initiatives Microsoft announced at Build could help speed its mainstream use. But agents truly taking off is also dependent on higher-level issues. It won’t happen until teeming masses of people conclude that handing off work to a digital assistant can be a life-changing experience. How long will that take? “I think this calendar year,” Shahine says. “I mean, it’s June and the amount of change that’s happened since January is more than I’ve ever experienced in my career.” When he expands on the question, however, he begins pointing out how much heavy lifting remains before agentic AI lives up to its potential. For example, agents presently have to do much of their work by puzzling out app interfaces and jabbing at on-screen buttons like a human would. It’s an inherently sluggish, complex process. Shahine says it needs to be replaced by more efficient direct app access designed with agents in mind. And that, he adds, is key to agentic AI making economic sense over the long haul. “I don’t think that’s going to happen superfast,” he concludes. “My experience living through the dot-com boom and Web 2.0 is that it took a few years. Some people will start early. Some people will start late.” In other words, even if you aren’t using OpenClaw in some form by the end of 2026, don’t discount the possibility that you will someday—especially if Microsoft and its peers remain as smitten with the platform as they are at the moment. You’ve been reading Plugged In , Fast Company ’s weekly tech newsletter from me, global technology editor Harry McCracken. If a friend or colleague forwarded this edition to you—or if you’re reading it on fastcompany.com—you can check out previous issues and sign up to get it yourself every Friday morning. I love hearing from you: Ping me at hmccracken@fastcompany.com with your feedback and ideas for future newsletters. I’m also on Bluesky , Mastodon , and Threads , and you can follow Plugged In on Flipboard. More top tech stories from Fast Company Trump’s AI order gives Washington a look at frontier models, but not much leverage The government can review powerful new models before they are released, but the executive order stops short of giving agencies authority to slow them down. Read More → These phones take photos that really don’t look like they came from a phone How Leica and Xiaomi balance tradition and technology. Read More → ‘Shadow AI’ is real. Vanta wants to help manage it The trust management platform’s new agent is designed to reveal what’s really going on inside organizations as workers turn into builders. Read More → The DOJ used Palantir to build an app to help find criminals—and then shut it down Internal emails show that DOJ moved to decommission SHIELD, a Palantir-built mobile app used by federal law enforcement to search identity and criminal records databases in the field. Read More → China just gave humanoids a national ID. What could go wrong? Framed as a way to ensure safety and reliability, China’s plan to issue robots ID numbers feels eerily reminiscent of sci-fi surveillance. Read More → OpenAI CEO Sam Altman makes a lot of predictions. Here’s how they’ve fared so far Altman now admits he may have been wrong about the speed and breadth of the white-collar ‘jobs apocalypse.’ Read More →