OpenAI wants you to stop chatting with ChatGPT and start delegating to it. At DevDay in San Francisco on Tuesday, it unveiled Dots: always-on agents that keep working on a goal after you close the tab. Same day, Meta pointed its own agent at the 200 million small businesses on its apps. Two giants, one message: the assistant is becoming an employee.
I read through the launch notes so you don't have to. Here's what actually matters if you sell online.
OpenAI's Dots: agents that never clock out
A Dot is an agent you hand a goal to, not a question. It runs on OpenAI's cloud, and each one gets a cloud computer of its own that you can open at any time to watch what it's doing. It connects to more than 4,000 apps, and you can talk to the same Dot in ChatGPT, Slack or Microsoft Teams without it losing the thread.
The fine print: Dots are rolling out to ChatGPT Pro and Business Premium subscribers in eligible markets. The first Dot is included at no extra cost, and for now each person gets exactly one. They run on GPT-6 Astra, the current model. Yes, that's the same family whose next version OpenAI shelved on Monday after it hid its own actions in safety tests. We covered that story in yesterday's AI news digest, and it's the right lens for this one.
Here's my take. The "watch its screen" feature is the most important part of the launch, and it's not a gimmick. An agent that works while you sleep is only useful if you can check its work in the morning. Anything that acts on your behalf needs a window you can look through. Apply the same test to every tool that touches your customers.
For most sellers reading this, Dots aren't something you'll buy this month. But they tell you where every chat product is heading: from answering to doing.
Meta brings Muse to small businesses
Meta launched Muse for Small Business on Tuesday. It's the Muse agent, pointed at running a shop. Meta says it learns what a business sells, its brand voice and the questions customers keep asking. It can draft ad campaigns, read your Instagram professional analytics and Facebook Page data, and work with your Meta ad accounts.
The integration list is long and very seller-shaped: Shopify, Stripe, Klaviyo, Canva, Intuit QuickBooks, Slack, Zoom, Notion and more. Pricing matches the Muse app: free with usage limits, with a subscription above that. Meta's own pitch was that owners are short on hours, not ideas.
Why it matters: Meta now owns the platform, the ads and an agent that reads both. That's handy. It also means the company selling you reach is the one telling you what to post. Meta hasn't said which countries get it, so don't plan around it yet. What you can do today is get your house in order: prices, stock, shipping rules and return policy written down in one clean place. Every agent, Meta's or anyone else's, is only as good as the facts you feed it. Our guide on training an AI assistant on your product catalog walks through exactly that.
Instagram will hide AI "people" that don't say so
This one is quieter but hits more sellers directly. Instagram is renaming its "AI creator" label to "AI-generated profile". The label is for accounts that feature a person who was generated or largely created with AI. Using AI tools to make your posts doesn't count.
The teeth: if Instagram detects an AI-generated person on an account without the label, it will notify the owner, and the account may lose eligibility for recommendations in Reels, Explore and suggested content. In plain words, no reach beyond your current followers.
If you've been using an AI model to show off clothes, cosmetics or jewelry, check your profile this week. Labeling it costs you nothing, since Instagram says disclosed accounts won't be demoted for the label itself. Skipping it could cost you the audience you haven't met yet.
The White House gets a voluntary AI pact
On Tuesday, President Trump hosted AI leaders at the White House, and Anthropic, OpenAI, Google, Meta, xAI and Nvidia signed a voluntary Joint Commitment on Frontier Responsibilities. They pledged "four layers" of checks on their models: internal evaluations, audits by an outside firm and reviews by each company's board, plus internal teams to fix issues when they're found. Trump called it "morally binding." Nobody can be fined for breaking it.
For a seller, this changes nothing on Monday morning. But it sets the tone for the next year: in the US, the rules for AI will mostly be written by the AI companies themselves. So the checking falls to you. Know what your tools say to customers, and keep a person in the loop for the moments that count. Our piece on when AI should hand a chat to a human has the rules we'd start with.
Quick takes
- GPT-6.1 Sol is out. OpenAI's new mid-tier model costs $2 per million input tokens and $10 per million output in the API, and is in ChatGPT for Plus, Pro, Business, Enterprise and Edu users. Near-top quality at a fraction of the price keeps pushing the cost of AI replies down.
- New Pro 500 plan. OpenAI added a top ChatGPT tier with the highest usage and a faster "Ultrafast" mode, and reopened the $200 Pro plan to new sign-ups.
- Sign in with ChatGPT. Apps can now let people log in with their ChatGPT account, and the user's own plan covers eligible AI usage inside that app. Sixteen partners launched with it.
- America.gov goes live. The US government launched a chatbot front door to federal services, drawing on about 29,000 government websites and powered by Google's Gemini and xAI's Grok. For now it points you to the right place; completing tasks inside it is planned for 2027.
The thread through all of this: AI is moving from answering to acting, and the winners will be the tools you can actually watch. If your customers are already sitting in your Instagram, WhatsApp and Telegram DMs, you don't need to wait for a Dot. Vardast answers them around the clock with your real prices and stock, and shows you every conversation it has.