🖥️ Claude Can Now Control Your Computer

Claude can now actually use your computer for you. From opening apps to filling in spreadsheets, it’s designed to handle everyday desktop tasks like a real coworker. It hints at a future where your AI doesn’t just assist, but were you actually hand over the keyboard. Let’s dive in!

In today’s insights:

LATEST DEVELOPMENTS

Evolving AI: Anthropic has given Claude the ability to directly operate a user's Mac desktop: opening apps, navigating browsers and filling out spreadsheets.

Anthropic has rolled out desktop control for Claude, letting the AI handle tasks users would normally do themselves at their computer. The feature is part of Claude Cowork and Claude Code and currently runs only on macOS. Claude doesn't jump straight to controlling the desktop , it first tries existing integrations like Slack or connected calendar apps. Only when no suitable interface exists does it take direct control, making it a last resort rather than the default.

The technology builds on Anthropic's acquisition of Vercept AI, a startup focused on AI-powered computer control. Its co-founder said the team shipped its first product less than four weeks after joining Anthropic. Alongside this, Anthropic launched "Dispatch," letting users access and control their machine remotely from any device. The feature is still a research preview and questions around data privacy and error rates remain open.

The desktop has become the new battleground. What makes Anthropic's move interesting is the approach: Claude only takes control when nothing else works, which is a more cautious entry than rivals who went all-in from the start. That caution might actually be smart, because reliability is still a genuine problem across the board. On browser automation benchmarks, Claude sits at 56% success rates compared to ChatGPT agent mode's 87%: the gap is real. But if Anthropic can close it while keeping the safety guardrails tight, the Vercept acquisition starts to look like a longer play than it first appeared. The race to build an AI that actually does your computer work, not just talks about it, is well underway. The question is who gets there reliably first.

Evolving AI: A 22-year-old bartender just landed a $1M TV deal… without ever acting.

Higgsfield AI built him a full series using nothing but his face; no cameras, no crew, no scripts. Episode one is already streaming.

Go watch the first episode. It's free and worth 20 minutes of your time.

Watch now →

Evolving AI: Tencent has brought the open-source OpenClaw AI agent into WeChat giving over a billion users direct access to task automation.

Tencent launched ClawBot on March 22 as part of a broader push into AI agents that have become a key battleground among China's tech companies. The tool turns WeChat into a direct interface for OpenClaw allowing users to automate tasks like sending emails and transferring files without leaving the app. WeChat already functions as a super app in China handling messaging, payments, and social media. Adding ClawBot as a native contact extends that reach by turning the platform into a hub for agent-driven automation. This follows Tencent's earlier launch of its own AI agent suite comprising QClaw for individual users, Lighthouse for developers and WorkBuddy for enterprises. Alibaba launched Wukong, an enterprise platform that coordinates multiple AI agents to handle complex business tasks. Baidu followed with a series of OpenClaw-based agents spanning desktop, cloud, mobile and smart-home devices.

China has already surpassed the US in OpenClaw adoption, and this WeChat move shows why that gap could widen fast. Analysts say whoever gets users into the habit of "asking AI first" will essentially lock in the next era of the internet, and with a billion people already living inside WeChat, Tencent has a shortcut nobody else really has. People are already using OpenClaw to run one-person companies, booking flights, and directing other bots around the clock, so this is not just about chat features. The cutthroat spending and speed are drawing comparisons to how China dominated solar panels and EVs, industries where aggressive domestic competition eventually created global players. The question is whether the West is paying close enough attention this time.

This is Viktor, an AI coworker with its own computer. It lives in Slack, connects to 3,000+ tools, and does all the work: writes the code, handles the API calls, and deploys a live app. One message. Real output.

Try Viktor free.

Evolving AI: AI-generated content has quietly entered the 2026 midterm campaign trail, and not everyone is being upfront about it.

AI has shown up across state local and federal elections, used to enhance speech turn politicians into cartoons and in one notable case mimic a rival's voice. In Massachusetts Republican primary candidate Brian Shortsleeve created an AI-generated radio ad designed to sound like Democratic Governor Maura Healey using her "voice" to say things she never actually said. The clip carried no explicit AI disclaimer.

The trend goes beyond one race. The National Republican Senatorial Committee released an AI-generated video of Texas Senate candidate James Talarico reading real tweets on sensitive political issues. Former New York Governor Andrew Cuomo's mayoral campaign also used AI in ads depicting criminals as supporters of his rival. Mark Jablonowski CEO of the progressive advertising firm DSPolitical said that anytime generative AI is used to create misleading messaging or imagery it becomes a serious problem. At the same time he believes most campaigns still want to do the right thing. The question now is where the lines are, legally and ethically.

The real issue here is not just one misleading ad. 26 states have already passed laws to regulate AI in political advertising and yet campaigns are still finding gaps to work through. Texas has one of the strictest deepfake laws in the country, but it only kicks in 30 days before the election, which is exactly how the NRSC released a convincing AI video of a Senate candidate months out and stayed in the clear. At the federal level, the REAL Political Advertisements Act never even made it to a vote. Voters heading into the most expensive midterm cycle in history are navigating a landscape where a candidate's voice or face can be fabricated in hours and the rules meant to stop it are full of holes. The bigger question is not whether AI will shape these elections, it already is. It's whether voters will be able to tell what's real by the time November arrives.

Google just opened a brand-new AI hub in London called Platform 37, and this video takes you inside. It’s where developers, startups, and companies come together to build and experiment with AI in real life. It gives you a clear look at how AI is actually being built and tested today, not just talked about.

QUICK HITS

🎶 US man pleads guilty to defrauding music streamers out of millions using AI.

🤖 Meta AI’s New Hyperagents Don’t Just Solve Tasks—They Rewrite the Rules of How They Learn.

🌀 Kansas City using AI to better prepare for natural disasters.

🧬 AI rebuilds molecules from exploding fragments.

 *partner link