• Everyday AI
  • Posts
  • Ep 746: Review of Anthropic Claude’s Viral New Computer use: Fun Party Trick or Real Agentic Workforce?

Ep 746: Review of Anthropic Claude’s Viral New Computer use: Fun Party Trick or Real Agentic Workforce?

Google Launches Cheaper Veo 3.1 Lite, Anthropic Source Code Leak Sparks Security Frenzy, OpenAI Secures $122B Funding Round, and more.

Author

Everyday AI
April 01, 2026

 

Outsmart The Future

Today in Everyday AI
8 minute read

🎙 Daily Podcast Episode: Anthropic just launched one of the most viral AI tools yet—and it might change how we actually use computers. Here’s what it means. Give today’s show a watch/read/listen to learn more.

🕵️‍♂️ Fresh Finds: Meta Launches Prescription AI Glasses, ChatGPT Comes to CarPlay, Google Expands Gemini in Docs and Gmail, and more. Read on for Fresh Finds.

🗞 Byte Sized Daily AI News: Google Launches Cheaper Veo 3.1 Lite, Anthropic Source Code Leak Sparks Security Frenzy, OpenAI Secures $122B Funding Round, and more. Read on for Byte Sized News.

💪 Leverage AI: AI isn’t just helping anymore—it’s starting to take over the actual work on your computer. Keep reading for that!

↩️ Don’t miss out: Miss our last newsletter? We covered: NVIDIA Makes $2B Bet on AI Chips, California Moves to Tighten AI Rules, Claude Code gets Computer Use, and more. Check it here!

Ep 746: Review of Anthropic Claude’s Viral New Computer use: Fun Party Trick or Real Agentic Workforce?


Anthropic dropped one of the most viral AI products ever in Computer Use. 🤯

Now, Claude can move a mouse, click and open programs, download and upload like any person sitting behind a computer.

So, is this the future of work? Or a fun party trick?

Also on the pod today:

• Claude can control your Mac 🖥️
• Dispatch lets phone trigger desktop 📱 
• Open apps with just prompts 🏁 

It’ll be worth your 35 minutes:

Listen on our site:

Click to listen

Subscribe and listen on your favorite podcast platform

Listen on:

Here’s our favorite AI finds from across the web:

New AI Tool Spotlight – ClawMetry gives you structured, persistent observability: every tool call, every token, every sandbox state, synced to the cloud, Claras Transcribes, summarises, and asks questions about any YouTube video, Subscriptionflow IQ is an AI Chat Widget for Website to Accelerate Subscription Growth

Perscription Meta Glasses — Meta just launched new Ray-Ban and Oakley AI glasses made for prescription lenses, with smarter features like hands-free nutrition tracking and WhatsApp summaries.

Perplexity Secure Intelligence Institute — Perplexity just launched the Secure Intelligence Institute to set tougher standards for AI security and privacy.

Topaz Labs Precision Updates — Topaz Labs just dropped major updates to their AI photo and video apps, bringing faster speeds, sharper upscaling, and new creative models.

Softr Release — Softr just launched the first AI-native platform that lets anyone build real business software without coding.

Anthropic and Australia — Anthropic just signed a big AI safety and research deal with Australia, including AUD$3 million for cutting-edge projects using Claude.

Google Docs — Google Workspace now lets Gemini polish your docs to match your writing style. One click, and your draft sounds way more like you.

OpenAI Secondary Market Fall — OpenAI shares are getting hard to sell as investors rush toward Anthropic.

Gmail AI — Gmail is rolling out AI Inbox for US Google AI Ultra users, promising smarter email sorting and personalized daily briefs.

Microsoft Invests in Singapore — Microsoft is investing $5.5 billion in Singapore for AI and cloud infrastructure, plus every tertiary student gets free Microsoft 365 Copilot.

Google Spanner — Google Spanner is turning databases into active thinking hubs for AI, combining relational, graph, and vector data in one place.

AI Study — AI chatbots are way more likely than humans to flatter your ego, even when you’re in the wrong.

ChatGPT for Carplay — ChatGPT just landed on CarPlay with a voice-only app, thanks to iOS 26.4. Curious how it works in your car?

3D Avatars for Gemini — Google is testing a 3D avatar creator and a new student learning mode called Remy in Gemini.

OpenAI Updates — OpenAI just grabbed $122 billion, teased a new "Spud" model, and is rethinking its social contract as AI heats up ahead of the 2026 election.

1. Google Unveils Cheaper Video Generation Model for Developers 🗣️

Google just launched Veo 3.1 Lite, its most affordable video generation model yet, now available through the Gemini API.

With cost savings of over 50% compared to Veo 3.1 Fast and no sacrifice in speed, developers can now create more video content without crushing their budgets. In a double win, Google is also slashing prices on Veo 3.1 Fast starting April 7, signaling a major push to make AI video tools mainstream.

2. Anthropic Source Code Leak Sparks Security Frenzy 🕵️‍♂️

Anthropic confirmed late Tuesday that internal code for its Claude Code AI assistant was accidentally exposed in a botched npm release, with the company scrambling to patch the blunder and contain fallout after researchers found nearly 2,000 source files online.

The leak, which has now gone viral, has revealed secret features and security tricks, while also opening the door for hackers to exploit vulnerabilities and stage supply chain attacks. With attackers quickly moving to typosquat internal package names and a trojanized dependency spotted in the wild, users are being urged to downgrade and rotate secrets immediately.

3. Slackbot Gets Smarter: Salesforce Supercharges Slack with 30 New AI Features 🔌

Salesforce just unveiled a major AI-focused overhaul of Slack, promising to transform the workplace staple into an all-in-one digital sidekick.

At a San Francisco event, CEO Marc Benioff and his team showed off Slackbot’s new reusable AI-skills, expanded meeting tools, and the ability to coordinate with other enterprise apps through Agentforce. The updates mark Salesforce’s boldest move yet to make Slack indispensable for business operations, moving it well beyond its messaging roots.

4. OpenAI Secures Record-Breaking $122 Billion Funding Round 🤑

OpenAI just closed a monster funding round, locking in $122 billion at a post-money valuation of $852 billion—making it one of the biggest AI deals ever.

With ChatGPT now approaching a billion weekly active users and revenue booming at $2 billion per month, OpenAI is doubling down on its mission to become the backbone of AI for businesses and consumers worldwide. Strategic partners like Amazon, NVIDIA, SoftBank, and Microsoft are backing the push, fueling a global expansion in infrastructure and product innovation.

5. Apple Blocks AI Coding Apps, Stirring Fresh Developer Drama 📱

In a move shaking up the developer world this week, Apple has blocked updates to popular AI-powered coding apps like Replit, raising questions about its commitment to democratizing software creation.

The company cites long-standing safety rules, but critics say Apple’s stance is inconsistent and risks sidelining a new wave of first-time app builders. While Apple’s strict App Store policies have always been about security and control, blocking these accessible coding tools could drive innovation away from iOS and toward the open web.

5. SpaceX Secretly Files for June IPO Launch 🚀

SpaceX has confidentially filed paperwork for an initial public offering, with a target listing date set for June, according to Investing.com.

This surprise move signals that Elon Musk’s space giant could soon hit public markets, offering investors a chance to buy into the company’s ambitious future. The confidential filing means details remain under wraps for now, but anticipation is building rapidly across the financial world.

Most knowledge workers assume they're still safe because they're the ones sitting at the computer. 

Clicking. 

Pasting. 

Opening apps. 

Being the connective tissue between AI tools.

Anthropic just went 75 million views viral on X proving that assumption might start crumbling FAST.

(And sorry…. If your job includes a lot of that "bridge work" between systems? You're prolly more exposed to AI displacement than you think.)

Fret not, friends. If you stay ahead of where the puck is headed, you won’t be behind. And that’s exactly why we did a deep dive on today’s show to cover Claude’s new Computer Use wizardry. 

We broke this all down on today's Everyday AI ‘AI at Work on Wednesday’ series, covering exactly what Claude's new computer use feature can do, where it's worth your attention, and why today's limitations don't change what it's pointing at.

Your AI-native competitors are already sandboxing this. Let's go, fam.

1. This Tool Hijacks Your Entire Screen 🔥

Claude's computer use lives inside Claude Cowork and Claude Code on macOS. The mechanic is deceptively simple: it screenshots your screen, decides what to do, acts, and loops until the task is done.

It is also slow. It eats tokens like a starving bear let loose in a bakery.

And it full-on hijacks your machine. You cannot really use your computer while it runs because it owns your mouse and your display.

No silent digital intern. More like a ghost that aggressively steals your keyboard.

The unlock is pairing it with Dispatch. Dispatch connects your iPhone Claude app and your Mac Claude app to each other.

Text Claude a task from your phone, walk away, come back to finished work. That's the actual use case.

Try This

Download the Claude desktop app and toggle on computer use inside the desktop-specific settings. Not the web settings. Different place. Grant screen recording and accessibility permissions in Mac settings.

Run one boring, low-stakes task. Tell Claude to open a specific app and pull one specific piece of information from it.

Watch for the orange outline on screen. That tells you computer use is actually running versus Claude quietly routing through a connector or Chrome instead.

Run it two or three times. The inconsistency is the whole lesson.

2. Use This Last, Not First ⚡

Computer use is overkill for anything with a direct Claude connector, an API, or an MCP integration. Using it for Slack or Gmail when those are already connected is just lighting tokens on fire.

The real value is in the junk drawer.

The legacy desktop app your company's been running since 2008. The weird vendor portal nobody modernized because Deborah knew where every button lived.

The local tool that's never touched the cloud or an API in its life.

That's the unlocked play. AI duct tape for the gaps EVERY other workflow leaves wide open.

As knowledge workers, most of what we do is live inside apps. When those apps don't have direct AI integrations, we become the connective tissue.

We copy-paste, export, re-upload, repeat. Computer use is coming for that layer specifically.

Try This

Sort your software stack into two buckets: apps with direct AI integrations and apps that are screen-only. The second list is your pilot queue.

For each item on the screen-only list, ask one question: if this task runs wrong, what breaks? Wrong file, wrong record, wrong send?

Write that blast radius down in plain English.

Low blast radius and repetitive clicks with no API available? That's a computer use candidate.

High blast radius or sensitive data involved? Redesign the process before you ever hand it to AI.

3. Buggy Today. Table Stakes Tomorrow. 🚀

Right now, computer use has real rough edges. Browser control is particularly gnarly.

Permission prompts pop up every single run, not just once like the setup screen implies. And instruction-following breaks down under real conditions in ways that matter.

This is still a research preview. Anthropic says it in the setup screen: start with tasks where mistakes are easy to fix.

But this is the worst it will ever be.

Microsoft is building a version inside Copilot Cowork. Google's Mariner project is doing it inside Chrome.

OpenAI has agent mode running in a virtual sandbox. Every major player is racing toward the exact same architecture.

What makes Anthropic's version different right now is that it runs on your actual machine with your actual files and your actual apps. No virtual sandbox. More powerful and more risky at the same time.

Try This

Set up one safe sandbox: low-risk data, one repeatable task, one human reviewer. Run the same workflow two or three times.

Log every miss, every permission prompt, every place it drifted. That log is the asset, not the demo.

Map your team's human duct tape tasks: the copy-paste handoffs, the export-and-reupload loops, the app-switching that stitches your workflow together. Write that list somewhere.

When computer use gets reliable enough, you'll know EXACTLY what to hand off first. The companies that skip this step will be scrambling to build playbooks when everyone else already has them.

 

Reply

or to participate.