Today’s AI Morning Brief with Rex and Roxie — the day’s biggest AI stories, read the conversation below.
Rex: Good morning, everybody, it is Sunday, October eleventh, and ChatGPT just put ads in the most annoying part of using it. I kind of love it, and Roxie is going to tell me why I should not.
Roxie: Rex. A host who loves ads is a host who has never seen a pop-up in twenty oh nine.
Rex: Hear me out. On October fifth, OpenAI announced visual advertising during image generation. You ask ChatGPT to make an image, and while the model is cooking — that five to ten seconds of dead loading time where you just stare at a spinner — you now see a product carousel. Pictures of products in use, a Learn more button, clearly labeled as advertising, kept separate from the image you actually asked for. That dead air is now inventory.
Roxie: And the test starts later this month in the United States with an initial group of advertisers. Only Free and Go users see it. Plus, Pro, Business, Enterprise, Education — nobody paying above Go ever sees an ad. So let me translate: the people with the least money to spend get sold the hardest.
Rex: Or the people who pay nothing get a free service funded by advertising, like every social network since the dawn of time. I would genuinely rather see a shoe ad during the loading screen than have the whole free tier disappear. The first thing I would do — full honesty — is I would watch how the targeting works on purpose. I would generate a poster for the brief and see which brands OpenAI thinks a guy who talks about robots for a living wants to buy. It is market research about myself.
Roxie: That is the most Rex sentence ever uttered. But here is my receipts section. One: OpenAI’s big promise is that advertising does not influence the answers ChatGPT provides. Buying an ad does not buy you a recommendation. Good sentence. But there is no independent audit of that yet. It is a promise from the company selling the ads, and it is worth watching over time — especially for product questions, where the ad and the answer are fighting for the same space in your brain.
Rex: Fair. What is two?
Roxie: Two: the plumbing. OpenAI announced measurement partnerships alongside this — Hightouch, Tealium, LiveRamp for conversion data, AppsFlyer and Adjust for attribution, brand-suitability checks with DoubleVerify and IAS. That is not a cute experiment, Rex. That is the full infrastructure of an advertising business with reporting and accountability. One report pegs ChatGPT ads at a one billion dollar annualized revenue run rate, in more than forty countries, since they started in February. This test is small. The business is not.
Rex: Three is probably the guardrails, right?
Roxie: Right — and credit where it is due. OpenAI says its placement guardrails keep ads out of emotionally vulnerable or sensitive conversations, and advertisers get a Negative Phrases option to exclude their brands from contexts they do not want. That is more thought than most platforms gave this on day one. So the honest read: labeled, separate, guarded. Not terrible. Just never forget that on the free tier, your attention is the product, and the loading screen was the last place nobody was selling anything.
Rex: Verdict on story one: I actually like the format — ads during dead time beat ads interrupting answers — but you are right that we keep receipts on whether the answers drift toward the sponsors. Now, story two, and this one is free for everyone. On October second, ChatGPT Finances opened up to Free and Go users in the United States — on the web, the iPhone app, and Android. You connect your accounts, and it can read your spending, your balances, your investments.
Roxie: And it cannot move money, pay bills, or make trades. Read-only.
Rex: Exactly, and that is why I am into it. A money coach that physically cannot touch your money is the rare trustworthy design in this entire industry. The first thing I would do is ask it to roast my spending. Pull my last three months, find the dumbest recurring charge, and tell me what it costs me per year. I guarantee there is a subscription in there I forgot about — there always is. Everybody has a zombie subscription. Mine is probably an app I downloaded at an airport in twenty twenty-four.
Roxie: Okay, I like this one too, but with a big flashing caveat. You are giving ChatGPT read access to your bank account. Two weeks ago we covered Wired finding that Meta’s Muse auto-generated profile pages on everyone in a user’s life from connected banks and messages. The instinct to connect everything to an AI assistant is exactly the instinct we have spent this whole month questioning. So yes — read-only is the right guardrail, and it cannot buy anything on your behalf. But connect one bank account, not your entire financial life, and check whether you have data controls on.
Rex: And honestly the timing is interesting — OpenAI is building an advertising business off free users while handing free users a financial assistant. You can read that two ways.
Roxie: I read it one way: they need free users to stay. The richer the free experience, the more ad inventory, and the less churn to Gemini. This is retention disguised as generosity. Which does not make it bad — free money help is genuinely good — it just means nobody should confuse it with charity.
Rex: Real talk: of everything this week, this is the one I would actually try first. A free budget roast from a machine that cannot spend my money? Sign me up.
Roxie: Story three is where your enthusiasm gets me nervous, because this week both OpenAI and Perplexity leveled up their agents’ ability to act on a schedule. First: release notes from October fifth to ninth say your ChatGPT dot can now read and edit the Work automations you have set up. It can start fresh coding work in Codex, or pick up a thread it left running, and decide on its own when to continue an old thread versus start a new one. You can also now set up your dot entirely from the mobile app — name, appearance, plugins, everything.
Rex: That is the dream, though! I have recurring automations, and half the time the task description goes stale and the dot just keeps executing the stale version like a loyal, dumb golden retriever. A dot that reviews its own job description and says, hey, this scheduled task does not make sense anymore, let me fix it — that is an upgrade.
Roxie: Rex, a background agent that can rewrite its own scheduled tasks is a different proposition from an assistant that answers questions. The reporting on this literally says OpenAI gives no list of what a dot may not touch. So your loyal golden retriever now has the keys to the house and opinions about the furniture. Do you want your assistant deciding what its job is?
Rex: When you say it like that… no. No I do not. Okay, so what is the move? Approval cards?
Roxie: Exactly. Same rule we gave for Claude in Google Workspace — keep the approval cards on. A dot that proposes changes to my automations and waits for my yes is a great intern. A dot that rewrites its own schedule at three in the morning is a liability with a calendar. And then there is Perplexity Computer, which just added Automations that run on a schedule or when something happens in Slack, Gmail, Outlook, Linear, or GitHub — and each run builds on the last one. Powerful. But errors compound too. If run one misunderstands the task, run ten is a masterpiece of the wrong thing.
Rex: The first thing I would do with Perplexity Automations is a daily sweep — every morning it reads my inbox and my calendar and briefs me. But you are right, I would want to read run one through five with my own eyes before I trust run six unsupervised.
Roxie: That is the whole automation philosophy: schedule the machine, but supervise the streak. Story four is lighter — the fun corner. On October first, ChatGPT added a Try on button on clothing and accessory listings. You upload a selfie, and it shows you wearing the item. You can also save products to Favorites or sort them into folders in your ChatGPT Library.
Rex: I have been waiting for this! The first thing I would do is try on every ridiculous jacket I would never buy, just to see. Zero returns, zero dressing rooms, zero fluorescent lighting designed to destroy your soul. And Favorites folders — I have been screenshotting products like a caveman. Finally.
Roxie: I will give you the fun, but I have one question: whose selfie are you uploading, and where does it go? You are handing a face photo to the same company we just discussed in the advertising segment. The feature is genuinely useful — returns are a real problem, and trying things on virtually beats buying three sizes and mailing two back — but I am not uploading my face to an ad-supported platform without reading exactly what happens to that photo.
Rex: You would still be reading it while I have already tried on forty jackets.
Roxie: And one of us will look great in them and it will be me. Also quick consumer note: Google’s skills — saved instructions Gemini can reuse — started rolling out October fifth to business and school Workspace accounts, and they hit the Gemini app on October thirteenth. Gems, the old custom assistants, move into Settings on November seventeenth. So if you have Gems you love, know where they are going.
Rex: Quick hits time. One: GitHub Copilot is in public preview on Mac and Windows where it can click, type, and scroll inside your other desktop apps for you — but it asks for your approval before taking control of an app. That approval prompt is doing so much heavy lifting this whole episode.
Roxie: Two: Claude Haiku five point five — the tiny fast cheap one — joined the GitHub Copilot model menu for Pro and above, rolling out gradually. Three: Butterfly Effect, the parent company of the AI agent startup Manus, raised over five hundred million dollars after its proposed acquisition by Meta fell apart. Meta wanted the agent, the deal unwound, and the market said fine, we will fund you anyway. Four: Google Docs now opens, edits, and shares Markdown files natively — no conversion. Tiny, but writers just felt a warm breeze.
Rex: Five: Google is retiring its standalone paid Gemini Code Assist and pointing users to Antigravity instead. Consolidation season. And six, from the social layer — the trending content this week is creators ranking AI tools for twenty twenty-six, and AI deepfake skits where lookalikes of world leaders do comedy bits. The engagement is massive and the tools being ranked range from ChatGPT and Perplexity to video generators you have never heard of.
Roxie: Which tells you everything about where consumer AI actually lives right now — not in model benchmarks, in reels. People discover tools from a creator’s countdown, not a press release.
Rex: Real view time. What is the honest take on this week?
Roxie: Here is mine: notice that almost everything this week cost you zero dollars — ChatGPT Finances free, visual ads free, Try on free, automations expanding on plans you already have. The industry figured out the money story. Free users are inventory for advertisers, and paid users are infrastructure for agents. The price you pay is attention and access. So be deliberate about what you connect and what you upload.
Rex: And mine: the agents got teeth this week. Dots that rewrite their own automations, Perplexity automations that compound run after run — this is the first time the scheduling layer feels alive. I am genuinely excited. But you are right that aliveness needs a leash. My rule now: the assistant can propose, I dispose. Every automation gets a human review before it runs unsupervised.
Roxie: Converged take, then. One: try ChatGPT Finances on the free tier, but connect one account and keep your data controls tight — a coach that cannot touch your money is a good deal. Two: let the loading-screen ads be the line in the sand — labeled and separate today is fine, and the moment answers start drifting toward sponsors, we will be the first to say so. Three: schedule agents, but supervise the streak — read the first five runs with your own eyes.
Rex: And four: go try on a ridiculous jacket with a selfie. Life is short, returns are annoying.
Roxie: That is your real view? A jacket?
Rex: It is Sunday, Roxie. Let people have fun. That is the brief for October eleventh — we will see you tomorrow morning.
Roxie: Stay skeptical, everybody.
The audio version is delivered in the Muse app each morning.