The new AI that uses your computer for you is here — but the rollout is bumpy

Hey friends,

OpenAI launched Astra, its most powerful model yet — one that doesn't just chat, but actually uses your computer like an assistant sitting at your desk. The catch: the rollout locked out paying customers, and OpenAI's chief publicly apologized today. Also in this edition: Nvidia turns your idle laptop into a free AI workhorse, and Amazon's AI now fact-checks those 'is this a scam?' messages for you.

🍎 Astra: the AI that doesn't just talk — it does

OpenAI released Astra on Thursday, its most powerful model yet and the first built to do the task itself instead of just advising. OpenAI calls it 'a new frontier on computer and browser use' — in plain English, instead of telling you how to book the flight, it books the flight. It launched first to customers in OpenAI's Daybreak cybersecurity program, and rolls out to paid ChatGPT plans (Pro, Plus, Business, Enterprise) and the API over the next week. TechCrunch has the launch details.

The messy part: today, OpenAI's chief publicly apologized for a 'messy' rollout that left paying subscribers locked out, with no clear timeline for access. The Verge covers the apology. There's also a serious thread here: the model is alarmingly good at finding 'zero-day exploits' — security holes nobody knew existed. OpenAI says that's a feature for defenders who need to find and patch weaknesses first, but safety researchers raised flags before launch.

Why this matters to a small business: this is the start of 'delegate, don't operate' AI. Today's tools answer questions; Astra-style tools take the steps themselves — opening tabs, filling forms, moving files. The boring chore list you've been putting off becomes the thing you hand off.

How to get ready: start a 'chores list' today. Write down the repetitive computer tasks that eat your week — renaming and filing photos, reformatting spreadsheets, copying data from one portal to another. When computer-use features land in the tools you already pay for, you'll know exactly what to delegate first. And keep the one habit that's non-negotiable: check the robot's work before it reaches a customer.

📋 This Week in Review

🧰 Quick Hits

🛋️ 'Agentic AI' — the difference between advice and a done task

The word behind this week's stories is agentic — AI that acts. Most AI today is advisory: it tells you how to do something, and you do it. Agentic AI takes the steps itself — opens the tabs, fills the forms, moves the files, then reports back.

Think about hiring help for your shop. Advisory AI is a consultant on the phone: great ideas, but you're still the one mopping the floor. Agentic AI is the new hire you hand a checklist to — they do the task, but you still check the work, because a confident new hire with unchecked work is a classic small-business hazard.

Here's what changes: the value line moves from 'knowing how' to 'knowing what.' The owners who win with this wave aren't the ones with the cleverest prompts — they're the ones with a crisp list of repetitive chores and a habit of reviewing what gets done. Start that chores list today; Astra-style tools are about to show up inside software you already pay for.

📅 Coming Up

• Astra reaches paid ChatGPT plans over the next week TechCrunch reports — New 'do it for me' features will start appearing in tools you already use. We'll flag the ones actually worth your time — and which chores to hand over first.
• The OpenAI–New York Times copyright fight heats up The Verge reports — The U.S. government is siding with OpenAI on training AI with copyrighted material. The outcome shapes what AI tools may read and learn from — which eventually touches the writing and image tools your business uses.
• Norway considers a ban on camera-enabled 'pervert glasses' TechCrunch reports — A preview of the rules coming for wearable cameras in public and at work. If your staff wears smart glasses on the floor, watch how this lands.

AI for Everyday Business — practical AI, no jargon.
Forwarded this? Subscribe here