Key Takeaways
- OpenAI will test picture ads inside ChatGPT image generation later in October — US only, free and Go plans, clearly labeled, with tools that track whether an ad led to a sale
- Meta and Sierra announced the Personal Agent Protocol so businesses can tell a customer's AI assistant from a person or a junk bot; spec due by the end of October, nothing live yet
- Claude for Google Workspace (beta, paid Claude plans) sits in a sidebar in Docs, Sheets, and Slides and edits the file in place — formulas, pivot tables, charts, and matching slides
- GPT-6 Intelligent UI rolled out to paying users Wednesday and free users Thursday: answers can include buttons, charts, forms, and calculators; OpenAI says ChatGPT reaches 1.2 billion people a week
- A finished-looking calculator is still a draft — check one answer by hand before anything goes in front of a customer
Week of October 5–11, 2026
Four things happened in AI between October 5th and the 11th that are worth five minutes of your time as someone who runs a business.
All four point the same way. AI is moving into the places you already work.
This is Last Week in AI — the signal, not the noise. Four stories operators need to understand, and what each one means for how you run your business.

Story 1: ChatGPT is adding picture ads
OpenAI announced it will test a new kind of ad in ChatGPT: picture ads that show up while ChatGPT is making an image for someone. The test starts later in October, in the US only, with a first group of advertisers.
OpenAI says the ads will be clearly labeled and kept separate from the image being made, and that ads do not change ChatGPT's answers. Only people on the free and lower-cost Go plans see ads. OpenAI also added tools that help advertisers see whether an ad led to a sale.
What it means for operators: ChatGPT is being built into an ad platform, piece by piece. New ad formats, plus the tracking that tells an advertiser whether the money worked. If what you sell looks good in a photo — a finished kitchen, a restored car, a new smile — this is a spot to keep an eye on. It is not open to everyone yet. (This continues the ChatGPT ads opening we covered in September, when the $50,000 minimum came down and small businesses became a material share of buyers.)
Story 2: AI that shops for your customers got its first rulebook
Meta and an AI customer service company called Sierra announced a shared standard for how personal AI assistants deal with businesses. They call it the Personal Agent Protocol. Walmart, Shopify, and Stripe are among the partners.
Sierra is run by Bret Taylor, who is also chairman of OpenAI. He told CNBC: "Companies will know when it's a personal agent versus an actual person." He compared it to signing into a website with your Google or Facebook account.
The first version is due by the end of October. Nothing is live yet, and no other businesses have signed on.
What it means for operators: right now, when a customer's AI assistant visits your website, you can't tell it apart from a person or a junk bot. This is the industry's first real attempt to fix that: the assistant says who it works for, and the business decides what it is allowed to do. There is nothing to do about it today. But if you book appointments or take orders online, this is the plumbing that will decide whether AI assistants can book with you.
Story 3: Claude moved into Google Docs, Sheets, and Slides
Anthropic released Claude for Google Workspace, in beta, for anyone on a paid Claude plan. It puts Claude in a sidebar next to the file you have open. It reads the doc, sheet, or slide deck and makes changes right in the file.
In Sheets, it can write formulas, build pivot tables and charts, and add new tabs. In Slides, it builds new slides that match the look of your deck. You can approve each edit before it lands. Claude can only open the files your Google sharing settings already let you open.
What it means for operators: until now, using AI on a document meant copying text into a chat window, then pasting the answer back. That step is going away. The AI is coming to the file instead. For most owners, the Sheets part matters most: formulas and pivot tables are where people usually get stuck. It needs a paid plan, and it is still a beta. If you would rather stay in a free chat and upload a file yourself, the spreadsheet questions guide still works on the free tier — with the same rule that these tools are not calculators.
Story 4: ChatGPT stopped answering only in text
OpenAI rolled out a new model, GPT-6, with what it calls Intelligent UI. Paying users got it Wednesday. Free users got it Thursday, worldwide.
ChatGPT answers can now include buttons you can tap, charts you can adjust, forms, and small calculators built for your question. One example OpenAI showed was a personal savings calculator. If you prefer plain text, you can turn the visuals down. OpenAI says ChatGPT reaches 1.2 billion people a week.
What it means for operators: the free tool most of your customers and staff use just changed. Ask it to compare two quotes or work out a payment plan, and it may hand you a working calculator instead of a paragraph. That is useful. It also means a wrong number can arrive looking finished and official. Check the math before anything goes in front of a customer. (Same failure mode as what AI is actually bad at: plausible and polished is not the same as true.)
Also this week
On Monday, the Wikimedia Foundation, which runs Wikipedia, said AI agents it believes OpenAI was running made edits to its sites without permission and sent millions of automated requests. OpenAI acknowledged the findings. Wikimedia's tech chief said the cleanup burden is falling on everyone, including smaller organizations.
Putting it together
ChatGPT is turning into an ad platform. Big companies started writing the rules for AI that shops for your customers. Claude moved into Google's office apps. And free ChatGPT started building tools instead of writing paragraphs.
AI stopped being a separate place you go. It is showing up inside the things you already use.
That makes it easier to start, because there is less to learn. It also means checking its work matters more, because the answers look more finished than ever.
— Aristotle Taylor, CEO & Co-Founder, Levron Labs
What to do this week
Next time you open ChatGPT, ask it to build a quick calculator for something you price every week. If it builds one, check one answer by hand before you trust the rest.
If the real gap is that quotes, follow-ups, and booking still depend on a person copying answers between tools, start with a free ops assessment — we'll map the gap between what your operation actually does and what your software thinks it does.
Sources
ChatGPT picture ads (Oct 5)
- BleepingComputer — announcement and measurement tools
- BetaNews — Oct 5 date and US test
- TechJuice — ads only on free and Go plans
Personal Agent Protocol (Oct 6)
- Tech Jack Solutions — announcement, partners, and end-of-October spec
- FourWeekMBA — Sierra's post and Taylor's CNBC quote
- DealsnBuy — not yet deployed or adopted
Claude for Google Workspace (Oct 6)
- Anthropic — official announcement
- Reworked — features and availability
- AlternativeTo — Sheets and Slides detail
GPT-6 and Intelligent UI (Oct 7 and 8)
- TechCrunch — launch and rollout days
- TechRepublic — rollout by plan
- BetaNews — free user rollout and 1.2 billion weekly users
Wikimedia (Oct 5)
