OpenClaw hit 250,000+ GitHub stars and 40,000 exposed machines within weeks of launch. The demand signal was undeniable. But so was the security disaster.
Anthropic studied every failure mode, and then it started shipping.
Four features in 30 days, each one a direct response to something OpenClaw proved people want and couldn't safely deploy at work.
How the Sprint Maps to OpenClaw
Anthropic announced Monday that Claude can now use your computer to complete tasks — opening apps, navigating browsers, filling spreadsheets, anything you'd do sitting at your desk. You can message it from your phone, and you come back to finish work on your desktop.
That's the core loop of what made OpenClaw famous. Anthropic just shipped it with guardrails for one user, on one machine.
The feature is currently in research preview, available to Claude Pro and Max subscribers on macOS. Claude will ask for permission before performing any actions on your computer.
Here's how the four-week feature sprint maps to OpenClaw's playbook:
OpenClaw gave you a text agent from WhatsApp that worked on your desktop. Anthropic shipped Dispatch on March 17 — a persistent conversation thread that follows you from your phone to your desktop, letting you assign Claude tasks from your iPhone and return to finished work.
OpenClaw used Discord and Telegram as control surfaces. Anthropic shipped Claude Code Channels on March 20. An MCP bridge to both.
OpenClaw gave you full OS access, browser control, and app manipulation. Anthropic shipped computer use in Cowork and Claude Code today.
OpenClaw had 100+ community skills with no review process. Anthropic launched a curated plugin marketplace with enterprise admin controls.
The one deliberate gap: OpenClaw runs a heartbeat daemon — always on, 24/7. Cowork requires your Mac to stay open with Claude Desktop running. For teams that need agents running 24/7 without depending on a laptop staying open, that gap matters.
The strategy is legible. Let open source take the arrows, ship the enterprise-safe version before anyone else can. OpenClaw proved 250,000 developers want to text an AI that controls their computer. OpenClaw also proved that desire produces one-click remote code executions, CrowdStrike threat advisories, and 20% malware rates in skill ecosystems.
Every constraint maps to a compliance checkbox. Connectors before computer use. Permission prompts before every action. Sandboxed execution. When enabled, Claude prioritizes connections to supported services like Google Workspace and Slack. If a connector isn't available, it will still execute the task — but with guardrails intact.
Dispatch is the missing piece that makes it real. Computer use without phone-to-desktop handoff is just a parlor trick. The combination of Dispatch and computer use creates the async loop, assign from your phone, come back to finished work, which makes this genuinely useful in a work context, not just a demo.
This does not kill OpenClaw. It addresses a different customer. Claude is still not a general-purpose, open, customizable, self-improving agent. Running a local Qwen model that SSHes into a DGX Spark, downloads new models on demand, and creates its own skills? That's never happening inside Anthropic's garden. Claude is for teams that want agents who work without asking their security team to look the other way. OpenClaw is for builders who want to own every layer. Both will exist, and they're solving different problems. The same is true for enterprise deployment. Claude is the agent. Managing a fleet of them across a team is a different category entirely.
What This Means For You:
If you've been sitting on the sideline waiting for AI agents to be safe enough to deploy at work, the excuse is gone.
Claude can now open apps on your computer, navigate a web browser, and fill in spreadsheets (with permission prompts, connector prioritization, and audit trails). The gap between "this is impressive" and "I can actually put this in front of my team" just got a lot smaller.
The question now is whether your workflows are mapped clearly enough to hand off. Start there. Know what you want Claude to own, what requires human judgment, and what the handoff looks like. The companies getting ahead of this are the ones that have already mapped which workflows are handoff-ready and which still need a human in the loop.
If you want help mapping that out, that's literally what we do.
Clutch. Just launched.
OpenClaw made it easy to get an agent running. Clutch makes it safe to run that agent at work.
Secure multi-agent deployment, built for teams that need more than a single-machine setup. We just launched.
Meta acquired the team behind Dreamer
Meta brought in the full Dreamer team under Chief AI Officer Alexandr Wang's Superintelligence Labs. Dreamer was building a personal agent OS for non-technical users. Meta's conviction in agents is stronger than ever, and the company is now building agents that are "truly personalized and always-on, with the ability to integrate across surfaces and wearables."OpenAI hired a senior Meta ad exec to run ChatGPT ad sales
Dave Dugan, former VP of global clients and agencies at Meta, joins OpenAI as VP of global ad solutions, reporting to COO Brad Lightcap. OpenAI faces a projected $15 billion cash burn in 2026, which explains why a CEO who once called advertising a "last resort" is now raiding Meta's ad bench.US Treasury launched an AI Innovation Series for financial services
The Treasury's AI Transformation Office kicked off a public-private initiative to help financial institutions scale AI while keeping governance frameworks current.
I've been running Claude's computer use feature since the announcement dropped this morning. It works. And it’s pretty cool.
The demo Anthropic posted — user running late, asks Claude to export a pitch deck as PDF and attach it to a meeting invite, comes back to it done — sums up the gist well.
The permission prompts add a beat of friction that most enterprise teams will actually appreciate. Your security team isn't going to love "it just does stuff" as a deployment model. "It asks before it acts" is a sentence that’s actually going to get budgets approved.
If you're on a Pro or Max plan on macOS, update both the desktop and mobile apps and try it today. The async loop — task from phone, finished on desktop — is the thing worth testing.
Haroon
PS - Computer use is impressive on one machine. Clutch is built for when you need ten agents running across a team, with the security and oversight to match.