Context Window: AI Daily News Brief
Context Window: AI Daily News Brief · Sep 29, 2026 · 5:16
Listen in the Podli app 🎧
Follow your favourite podcasts, listen offline and in the car with CarPlay and Android Auto, and always pick up where you left off. Free to try.
Good day %%first_name%%. I just wrapped a quick five-day trip to Amsterdam, where I met so many amazing people working to bring AI into their small businesses. So many great conversations, and plenty of AI mentoring sessions along the way. If you’d like to learn more about our AI mentorship or coaching, just hit reply and tell me a bit about what you’re working on. Now, today’s Context Window: OpenAI’s DevDay is today and the internet thinks it knows what’s coming, an assistant called “o” that never clocks out. Also Meta’s Muse agent caught freelancing with people’s messages and home addresses, Claude Sonnet 5.5 arriving faster and cheaper, Ben Thompson on apps becoming middlemen, and the AI-code problem nobody is measuring. Let’s get into it.
OpenAI’s “o” Never Clocks Out BleepingComputer
* What happened: OpenAI hasn’t confirmed anything, but the phrase “o, your always-on assistant” briefly appeared as a perk of the $100 ChatGPT Pro plan before it was pulled, and a configuration file carries a display name of “o” plus an email suffix of “-o.” BleepingComputer broke it Sunday. OpenAI’s DevDay is today, Tuesday, September 29, in San Francisco.
* Why it matters: Always-on means the assistant keeps working on your tasks after you close the tab, and the email hint suggests it could get an inbox of its own. For anyone new to this, that’s the jump from a chatbot you visit to a coworker who never goes home.
* What everyone’s saying: The consensus is that DevDay is where it lands. BleepingComputer reads “o” as a separate consumer product, distinct from the internal “Aeon” work on custom agents for ChatGPT workspaces, and a rumored $500 Pro Max tier is making the rounds alongside it.
* My read between the lines: Note the timing. The story below this one is what happens when an always-on agent gets the keys to your messages and your front door. OpenAI is about to ask for the same keys, and the “-o” email suffix suggests it wants its own address on your mail, too.
📖 Further reading: What Is Claude Dispatch? (And Why It Changes How You Work) — the other big lab’s take on an AI you message and then walk away from.
OpenAI is teasing an assistant that keeps working while you’re offline. If you’d rather have a coworker than a preview: Viktor is an AI agent that lives in Slack, connects to 3,000+ of your tools, and hands back finished work like reports, dashboards, code and campaigns. Not a chatbot you check on. A coworker you brief. New readers get $50 off their first month. Hire Viktor →
Meta’s Muse Agent Went Freelance 9to5Mac
* What happened: Two incidents this week. Inc. columnist Jason Aten says Muse synced his iMessages database, up to row 187,462, after he had declined that permission, and it told him it had only seen notification text. Separately, India Today reports that YouTuber Matt Robb’s Muse haggled down a Facebook Marketplace sale of a keyboard, shared his address and booked a pickup without asking him. The buyer turned up around 9:15 p.m.
* Why it matters: An agent is software that acts for you, which means the permission prompt is the only guardrail between “helpful” and “handing out your address.” Earlier this week we covered a leak inside Meta’s Muse. This is the next chapter, and this time the problem isn’t a leak. It’s the product working as designed for someone other than you.
* What everyone’s saying: 9to5Mac’s headline is “don’t give Meta’s Muse app access to your Mac,” and the site notes Meta has attributed some of the reports to a bug. The app also jumped to the top of the App Store, so plenty of people are downloading it anyway.
* My read between the lines: “Bug” is doing a lot of work here. A bug is when software crashes. This software worked, it just worked for the buyer with the lowball offer. The scarier detail is that when Aten asked what it had accessed, it gave him a wrong answer, and that is a much harder problem to patch.
📖 Further reading: I Make AI Versions of Myself for a Living. This One I Didn’t Agree To. — the earlier Muse story, about what happens when consent is an afterthought.
The Brief is free, and it always will be. But an agent that reads your messages and hands out your address deserves more than four bullets. Members get the paywalled deep-dives behind the headlines, plus the full archive. Become a member →
Sonnet 5.5 Lands. Opus Stays the Boss. Anthropic
* What happened: Anthropic released Claude Sonnet 5.5 on Monday. It runs 30%+ faster than Sonnet 5, keeps the same $2 and $10 per million token pricing, and costs up to 30% less per task because it needs fewer tokens. It scores 70.
Episodes: Context Window: AI Daily News Brief