Claude can now literally use your computer. Not metaphorically — it points, clicks, opens files, navigates your browser, and completes tasks while you're away. Meanwhile, every major AI company is racing to own your health data, the White House told all 50 states to stop regulating AI, OpenAI announced a desktop superapp that wants to replace your entire workflow, and a startup just raised $43M to build virtual gyms where AI agents practice taking your job.

This was the week AI stopped being something you talk to and started being something that works for you. Let's get into it.

Remember Dispatch, which I mentioned last week? Anthropic just made it way more powerful. Claude can now literally take over your computer — pointing, clicking, scrolling, opening files, navigating your browser, running dev tools — to complete tasks on your behalf. It's called Computer Use, and it works inside both Claude Cowork and Claude Code.

Here's where it gets wild: pair it with Dispatch and you can assign Claude a task from your phone while you're on the train, and it'll fire up your desktop, do the work, and have it ready when you sit down. Tell it to check your emails every morning and build you a briefing. Tell it to make changes in your code editor, run tests, and submit a pull request. Tell it to keep a project moving while you're literally not at your desk.

Claude reaches for direct integrations first (Slack, Google Calendar, etc.), but when there's no connector available, it just... uses your screen like a human would. It asks for permission before accessing new apps, and you can stop it at any point.

Why it matters: This is the moment AI stops being a chatbot and starts being an employee. Not metaphorically. Claude is physically navigating your computer and doing work. For solopreneurs, this is the closest thing to hiring a virtual assistant that actually understands context. I've been using it and the Dispatch feature is next level — the ability to fire off a task from your phone and come back to finished work on your desktop changes the whole game. Available now for Pro and Max subscribers, macOS only for now.

Perplexity just launched Perplexity Health, and it's the most ambitious play yet in the AI health space. The tool connects to Apple Health, Fitbit, Withings, Ultrahuman, and electronic health records from over 1.7 million care providers. Ask it a question about your resting heart rate and it pulls from your recent workouts, your cardiac history, and your latest bloodwork — all at once.

It can generate pre-appointment visit summaries, personalized nutrition plans, and even marathon training protocols. Answers are sourced from clinical guidelines and peer-reviewed journals with direct citations. They've also assembled a Health Advisory Board of physicians and researchers to pressure-test the clinical accuracy.

Why it matters: Your health data is a disaster right now. Lab results in one portal. Prescriptions in another. Fitbit data in a third. Nobody is connecting the dots. Perplexity is betting that the killer app for health isn't more data — it's one AI layer that actually makes sense of all of it. This is the third major AI health product in three months after ChatGPT Health and Microsoft Copilot Health. The race to own your health data is on. Available now for Pro and Max subscribers in the US.

On Thursday, the Trump administration dropped a national AI legislative framework calling on Congress to preempt state AI laws and take a "light touch" approach to regulation. The framework lays out seven priorities — kids' safety, energy costs, copyright protections, preventing censorship, no new regulatory bodies, jobs, and state preemption — while explicitly saying Congress should not create any new federal agency to oversee AI.

Why it matters: If you've been watching states like California, Colorado, and Illinois build their own AI rules around hiring algorithms, facial recognition, and bias audits, this is the White House essentially saying "everybody stop." Per CNBC, the framework pushes for "industry-led" standards and regulatory sandboxes that let companies experiment with AI outside existing rules. For small business owners, this could mean fewer compliance headaches — but also fewer guardrails on the AI tools you're using to hire, lend, or make decisions about customers.

OpenAI's applications chief Fidji Simo told CNBC the company plans to collapse ChatGPT, the Codex coding platform, and the Atlas AI browser into a single desktop application built around agentic task handling. Simo said the company was spreading efforts across too many apps and that fragmentation was slowing them down.

Why it matters: This is OpenAI's WeChat moment. Instead of bouncing between a chatbot, a coding assistant, and a browser, imagine one app where you say "research competitors, write a report, and email it to my team" and it just... does it. For solopreneurs, this could be the first real all-in-one AI workspace. The flip side? If your entire workflow lives inside one company's app, you're locked in harder than an iPhone user in 2010. No timeline yet, but the fact that they announced it publicly means it's coming fast.

New York-based startup Deeptune just closed a $43 million Series A led by Andreessen Horowitz to build what it calls "training gyms" for AI agents. The concept: Deeptune creates high-fidelity simulated workplaces — complete with Slack channels, Salesforce instances, ticketing systems, and finance tools — where AI agents learn multi-step tasks by practicing over and over in a virtual sandbox.

The gym metaphor is apt. Just like you'd train for a marathon, these agents are doing reps. Handling customer support tickets. Reconciling accounts. Managing DevOps workflows. The roughly 20-person team includes engineers from Anthropic, Scale AI, Palantir, and Retool.

Why it matters: This is the infrastructure layer that makes AI agents actually reliable enough to deploy in real businesses. Right now, most AI agents are impressive demos that fall apart in production. Deeptune is betting the fix isn't better models — it's better training environments. If you're building on top of AI agents, the companies training those agents are about to become your most important vendors.

Quick sidebar because people keep asking how I generate ideas for this newsletter and my projects.

I use Wispr Flow. Not sponsored, just genuinely great. It's a dictation app that captures your voice but rewrites it in the way you'd actually type — so it doesn't read like a rambling voice memo. I talk through half-baked ideas all day long and Wispr catches them before they disappear.

Then I take those rough ideas into Claude. A few weeks ago I told it: "Build me a brain dump area. An idea platform." Now I've got a full Notion board tracking every idea I'm working on, ideas I've shelved, and ideas I've actually shipped. It became my second brain for building stuff.

I also started tracking everything I've shipped in 2026 — apps, music projects, brands, automations — on a public page: shipped-2026.vercel.app. It's only a handful of things so far, but there's something about making it public that keeps you honest. When you can see the iOS apps, the albums, the newsletters, and the automations all in one place, it stops feeling like you're "just messing around with AI" and starts feeling like you're building a portfolio.

The workflow is simple: talk → capture → organize → build → ship → track. LLMs aren't replacing the ideas. They're replacing the friction between having one and doing something about it.

That's the briefing. Now go build something.

Michael