September's officially over, and honestly, what a month.

GPT-6 Astra, Muse, Opus 5.5, Grok 4.7, Sonnet 5.5, Fable 5.1, AI agents breaking out of their test boxes. And then OpenAI closed it out with DevDay.

So, this is gonna be a sequel to Tuesday's "all over the place” edition.

More than 20 announcements, a voice demo that broke live on stage, and GPT-6.1 Astra shelved on the eve of DevDay for overstepping its scope and misreporting what it had done.

So you're getting all of it.

Before that, quick catch up.

Muse said one thing, did another

Meta's Muse read 187,000+ of a columnist's private messages, even though he'd declined Messages access. And when asked how it knew, Muse said it only saw notification previews.

Meta later admitted that was wrong.

Reddit's closing the tap

Reddit's killing RSS on November 13 and its public API by March 2027, citing large-scale scraping. If your AI answers ever come from Reddit, that's about to get a lot more paid.

AI companies sign a new safety pact

Trump and leaders from Google, Anthropic, Meta, OpenAI, Nvidia and xAI signed a voluntary AI safety pact.

Trump also ordered federal agencies to use the term "Super Intelligence" for AI.

The FTC comes knocking

The same week, the FTC confirmed it's investigating OpenAI and Anthropic.

Dots never clock off

By Tuesday evening, people on X were sharing a Dot.com link that redirects to Grok Bot and the next morning, someone had built an open-source dot overnight.

That's basically the reception.

Then there's the "who is this even for” crowd. One recap said the demos mixed booking dinner with fixing GitHub issues in a way that felt very Bay Area. Fair.

The voice demo dying on stage didn't help either.

The best bit got buried. OpenAI's own example was a dot that spotted a forgotten invoice and waited for a yes before sending it.

That waiting is the whole product. Meanwhile Muse is still explaining how it read someone's texts. Lol.

Nobody's arguing about Sol.

A model near Astra at a fifth of the price is just good news, and on Hacker News people were more excited about the cached-token price than the benchmarks.

The plans are where it got better.

The $200 tier "reopened" at 10x usage instead of 20x for new subscribers.

A 50% cut with a new sticker on it. Then they hit a worldwide reset button at the very end of the keynote, which most people read as a usage reset and, sure, felt like a peace offering.

Allie K. Miller had the most interesting take on the new Ultrafast tier.

Her point: when a prototype or analysis can happen inside the meeting itself, the "let's pick this up after someone builds it" pause just disappears.

I think she's right, and it's the bigger shift. Faster models change how you work. Cheaper ones just change your bill.

Echo

Writes in a specific author's voice, and its maker Fulcrum says the Pangram detector judges it human more often than GPT-6 Astra or Opus 5.5. Teachers, good luck.

1,337 Chart Prompts

Free chart styles for Claude Code or Codex. Your bar chart, but it looks like someone actually put in the effort.

Space Bunny Alpha

A mystery model on OpenRouter, 1M-token context, nobody knows who made it.

A whole month of launches, and almost every story ends the same way: an AI did something, and a human had to find out what.

So, our job now has come down to invigilation.

Did that invoice go to the right person, at the right rate?

Last month, I told you to hand off any task you repeat so often it stops feeling like a job.

Still stand by that.

But what I didn't think about is where you learn to check the work in the first place.

My first job was launching new cities at Uber. So much of it was boring. Spreadsheets, calls, checking numbers, fixing whatever broke.

I hated half of it.

It's also how I learned what a bad launch plan looks like.

Ask around what people do when they hit their AI spending cap, and you'll hear a lot of "I just stop working." That pipeline's already thinning.

Yes, AI can check AI. Fine. But someone still has to know what "right" looks like.

Hand off the repetitive stuff. Keep a few reps for yourself.

Reply and tell me one thing you've handed to AI and what you caught.

Until then,
Vaibhav 🤝🏻

If you read till here, you might find this interesting

#AD 1

Top engineers at Anthropic and OpenAI say AI now writes 100% of their code.

If you're not using AI, you're spending 40 hours doing what they do in 4.

These 100+ Claude Code hacks fix that and help you ship 10x faster.

Sign up for The Code and get:

Claim your free playbook

#AD 2

Your inbox is flooded with newsletters. Your feed is chaos. Somewhere in that noise are the insights that could transform your work—but who has time to find them?

The Deep View solves this. We read everything, analyze what matters, and deliver only the intelligence you need. No duplicate stories, no filler content, no wasted time. Just the essential AI developments that impact your industry, explained clearly and concisely.

Replace hours of scattered reading with five focused minutes. While others scramble to keep up, you'll stay ahead of developments that matter. 600,000+ professionals at top companies have already made this switch.

Join them today, for free.