Why Microsoft Is Betting on OpenClaw

You’re reading an issue of “The AI Economy,” my newsletter exploring the forces shaping the AI era—tracking how AI is rewriting business, work, technology, and culture. Subscribe to get expert insights and curated updates delivered straight to your inbox.

IN THIS ISSUE: Microsoft is making its boldest agentic bet yet, building a new team to bring the OpenClaw framework into Microsoft 365 Copilot — a move that could reshape how hundreds of millions of office workers interact with AI. But as enterprises race to deploy AI agents, a quieter crisis is building in the background: token costs are spiraling out of control, and most companies have no visibility into where the money is going. Ramp thinks it has the answer.

The Prompt

When it comes to the agentic era, Microsoft is all gas, no brakes. Despite its struggles to get enterprise customers to pay for Copilot, the company is continuing to reimagine how work gets done. In this pursuit to transform every business into so-called Frontier Firms, Microsoft is betting that the next big thing lies with the OpenClaw framework. Recently, it created a new team led by Omar Shahine, who previously led Microsoft Word, to build what he calls “a new generation of proactive assistants, ones that lighten your load by taking on tasks end-to-end, and that they can also step in proactively when they can help.”

Created by developer Peter Steinberger—who has since joined OpenAI—OpenClaw is one of the fastest-growing open-source AI agent projects. It functions as a virtual assistant that can run on your machine and do things on your behalf, including managing files, sending emails, browsing the web, and automating workflows across apps you already use. It’s the vision of agents taking control of the wheel at work—but its rise is being shadowed by security concerns.

These risks are one reason for the growing number of alternative frameworks. For example, after seeing that his OpenClaw agent had downloaded all his WhatsApp messages insecurely, developer Gavriel Cohen created NanoClaw. However, while there are at least four competing technologies, OpenClaw has currently gained the most traction. Its GitHub repository has more than 354,000 stars—suggesting a high degree of confidence from the developer community—and has been forked more than 70,000 times. There are now nearly 50,000 OpenClaw-related repos on GitHub, far more than any of the other “Claw families.”

Skill development is also accelerating, with more than 44,000 skills listed on ClawHub as of April 2026.

The enterprise world is paying attention. Tencent launched its own OpenClaw product suite last week, while Alibaba Cloud, Moonshot, and Xiaomi have released supported apps. Nvidia went further and built NemoClaw—an enterprise-grade security stack on top of OpenClaw—with Adobe, IBM’s Red Hat, and Box expressing interest. Salesforce, which is separately pushing deeper into autonomous agents, has acknowledged parallels between OpenClaw’s architecture and its own development roadmap.

Microsoft hasn’t officially revealed what Shahine’s team has built yet, though the earliest preview may come at Build 2026, set for June in San Francisco. What’s clear is that Microsoft already has the foundation in place. M365 Copilot supports agents that let organizations quickly build custom apps and workflows, Copilot Search provides a universal, AI-powered search layer that pulls from both Microsoft and third-party apps, and its Researcher agent now uses multiple models simultaneously to improve accuracy before a response ever reaches the user. And Microsoft brought the technology behind Anthropic’s Claude Cowork into Copilot, enabling long-running, multi-step work that executes across apps and files over time. Sound familiar?

You can start using OpenClaw in some Microsoft apps. In fact, there’s a “fully integrated” plugin for Microsoft Teams, enabling you to talk with your OpenClaw agent via direct messages, group chats, or channels.

“People are hungry for this. Not another chatbot. Not another tool that helps when you remember to ask. An always-on agent that works on your behalf, 24/7, with real access to your real life,” Shahine writes in a blog post.

His team, which he calls Ocean 11, is deliberately small, though each one is “a force multiplier.” Everyone is a builder and possesses a founder mindset. Shahine explains that their job is not to write features, but “build and refine the system that writes features.” Success for the team means developing a team of agents—not chatbots that respond when prompted—that operates 24/7/365 within Microsoft 365. They’ll continuously monitor what you’re doing, help you plan your day, assist you with all those emails, and tackle action items on your behalf.

Regardless of its risks, OpenClaw appears to be the future of the agentic enterprise. If Microsoft gets this right, OpenClaw stops being a developer curiosity and becomes the infrastructure layer for how hundreds of millions of office workers interact with AI. That’s a different kind of scale than anything OpenClaw’s community has imagined. The security concerns are real, and Microsoft will have to solve them before IT admins and business leaders fully buy in. And the company does have the infrastructure and software to mitigate risks and handle access permissioning at scale, especially for agents. But if everything goes according to plan, Microsoft 365 Copilot may be about to become something meaningfully more powerful than it is today.

Want to go deeper on OpenClaw? ClawBeat.co tracks everything — sourced news, research papers, events, and videos — all in one place.

A Closer Look

Tokens have become the currency of the AI era—and like any currency, they’re easy to spend and hard to track. Having too few can break workflows, and buying more can lead to a costly surprise at the end of the month. Data from Ramp shows that the average monthly spend on tokens has surged 13-fold. The figure highlights that businesses’ token usage has spiraled out of control — and until now, there’s been no effective way to manage the costs.

Enter Ramp’s AI Spend Intelligence, a tool that pulls token usage data from Anthropic, OpenAI, and OpenRouter to give administrators a clearer picture of where AI spend is going and who’s driving it, putting an end to the Wild West token spree. After all, companies are already monitoring transactions involving fiat currency and also how much is being spent on infrastructure like hosting, so why not also what’s being paid out to run AI workflows?

According to Ramp, “No tool in the market connects token-level usage, invoice data, and card-level spend.”

Knowing how many tokens you’ve consumed is one thing. Knowing what they cost before the bill arrives is another. Ramp’s new tool gives finance and IT teams the visibility to budget more accurately and avoid the end-of-month shock that’s become a fixture of enterprise AI spending.

Every AI interaction burns tokens, from asking a virtual assistant to summarize meeting transcripts or draft emails, to using Claude Code or Codex to build an app, to deploying an agent that resolves customer support tickets. They’re so valuable that some think people will soon receive not just monetary salaries but also compensation in the form of tokens. 

Ramp’s AI Spend Intelligence is currently available through its early access program.

Today’s Visual Snapshot

Data compiled by investment firm Andreessen Horowitz (A16z) via Crunchbase shows global VC funding to AI companies spiking in Q1 2026. The $242 billion figure represents roughly a 4x increase from the prior quarter and nearly 4x the previous single-quarter peak of $66 billion a year prior. 

Something to keep in mind: Q1 2026 was the quarter that included OpenAI’s $122 billion raise, as well as Anthropic’s $30 billion Series G, xAI’s $20 billion Series E, and Waymo’s $16 billion Series D. That’s $188 billion in investments by four companies! Without those, venture capital spending in the first three months probably would have looked similar to past quarters.

The bigger picture: The $242 billion is somewhat misleading in isolation. Nearly two-thirds of it flowed to just four companies whose capital needs look less like traditional venture bets and more like civilization-scale infrastructure financing. Still, the number signals something real: AI now commands 80 percent of all global venture activity—which coincidentally was the largest such quarter in history—, and three of those four giants are eyeing public markets before year’s end.

Quote This

“The dangers of getting this wrong are obvious, but if we get it right, there is a real opportunity to create a fundamentally more secure internet and world than we had before the advent of AI-powered cyber capabilities.”

— Anthropic CEO Dario Amodei on X, explaining the growing cyber capabilities of AI models and why his company is limiting the availability of its Mythos model to those enrolled in its Glasswing project.

This Week’s AI News

Industry Trends

Generative AI and Applications

Enterprise and Workforce

Retail and Commerce

Business, Marketing, and Funding

Science and Breakthroughs

Media and Entertainment

Policy, Safety, and Misinformation

Opinions and Analysis

End Output

💡 🤝 Did you like what you read in “The AI Economy?” Share it with your coworkers, friends, and family! Connect with me on LinkedIn and check out my blog for more insights and thoughts on business and technology.

🕵️‍♂️ 📨 Got a great idea for a story or have a juicy scoop? Feel free to shoot me a message here, or reach me securely on Signal at kenyeung.888.

✍️ 🚀 I also help companies translate innovation into storytelling that resonates. If your team needs a seasoned journalist to craft your next big narrative, I’m available for writing and content strategy projects.

🙏 📬 Thanks for reading, and be sure to subscribe to receive future editions. Stay curious!