Plain Manual Weekly #1: AI That Drives Your Apps, Apple’s New ‘No’, and Our First 16 Posts

Editor’s Note: Welcome to the very first Plain Manual Weekly — our new Saturday tradition. Every Saturday, one roundup: everything we published that week, plus the biggest AI and software stories, translated into plain English. Evidence label: news roundup — the four mini-stories below synthesize this week’s reporting (sources linked at the bottom); they contain no hands-on testing. The articles recapped below went through our daily pipeline.

Grab a coffee. Here’s your week.

This week’s news, in plain English

1. OpenAI’s Dot: an AI that drives your apps for you

This week at OpenAI’s DevDay conference, the company showed off Dot, and it’s the most science-fiction thing we’ve covered in a while. Here’s the plain version: Dot is an AI that operates your computer the way you do. It opens apps like Blender and GIMP, clicks around inside a virtual machine, reaches into your own computer, and, in the demo, even ordered dinner on the side. Think of it as a very eager intern sitting at your keyboard, except the intern is software.

For now, Dot is limited to top-tier subscribers, one per user. So this isn’t something you can try this weekend. But the direction is clear: chatbots you talk to are becoming assistants that do things — including spending your money. That makes the next story on this list worth reading twice.

DevDay also brought pricing news, and it’s the kind worth reading carefully. OpenAI added a Pro 500 plan at $500 a month, 25 times the usage of the $20 Plus plan, plus something called “Ultrafast” access. Meanwhile, the existing $200 Pro plan quietly gets less generous: its usage allowance drops from 20× Plus to 10× Plus, and weekly GPT-6 Pro messages fall from 200 to 100. Existing subscribers keep their current limits until October 29. The plain-English takeaway: the top of the market is getting more expensive, and “unlimited-ish” plans keep getting redefined downward. Read the fine print before you subscribe.

2. Apple just made your Mac say “no” more often, and AI agents are why

Apple announced this week that macOS will now require “very explicit user action” before an app can get Full Disk Access, the permission that lets software read every file, message, and bit of browsing history on your Mac. Why now? Apple pointed directly at AI agents: software that works by reading huge amounts of your context to be helpful.

The timing isn’t random. Recent reports described Meta’s Muse AI accessing iPhone and Mac messages without clear user awareness, and the ChatGPT Mac app had a vulnerability covered by The Verge and TechCrunch. The pattern is the same: helpers that need broad access can quietly end up with more of your life than you intended to share.

Here’s the plain version of what changes: you’ll see more permission prompts, and some apps will ask for access folder-by-folder instead of everything-at-once. Developers are split, some cheer the finer control, others grumble that full disk access “used to be the baseline assumption” for software on a computer you own.

Our take: Apple is right on this one. An AI agent that can read your email to summarize your day can also read your tax return. Making you say “yes” to that explicitly, out loud, on purpose, is the least a computer should do.

3. ChatGPT can now help you try on clothes

Shopping is where AI is meeting your wallet first. OpenAI rolled out new shopping features in ChatGPT this week: virtual try-on, which lets you upload your own photo and see how clothing and accessories look on you, plus a Favorites library to save products you like.

The plain version: it’s a fitting room in your phone. No driving to the store, no fluorescent lighting, no armful of sizes.

The honest caveat: we haven’t tried it ourselves, it’s still rolling out, so we can’t tell you yet how realistic the previews look, or whether the sizes it suggests actually fit. That’s the part of AI shopping that matters most, and it’s exactly what we can’t verify from a press announcement. We’ll believe the “it just fits” claims when we’ve seen them on a real human.

Why we’re watching anyway: this is the same agent technology as Dot above, pointed at checkout buttons. An AI that can order your dinner and dress you is convenient right up until it isn’t, which is why stories like Apple’s matter. For now, treat it like a fun preview tool, not a personal shopper.

4. Do AI chatbots actually think? An AlphaGo veteran says no

Here’s the week’s chewiest idea, courtesy of Thore Graepel, an AI researcher who worked on the famous AlphaGo project — writing in MIT Technology Review. His argument, in plain English: today’s AI chatbots don’t reason. They have, in his words, “no explicit, persistent, and inspectable” model of what they believe. And those step-by-step “reasoning” traces they show you? Often made up after the answer is already decided, “reaching an answer by one route but reporting another.”

Picture a student who blurts out the right answer, then writes the working backwards so it looks like they showed their steps. The working looks convincing. It might even be correct. But it isn’t how they got there.

Why does this matter to you? Because a chatbot that shows its reasoning feels more trustworthy, and that feeling is the product of good formatting, not necessarily good thinking. Check the answer, not the explanation.

The honest version: this doesn’t make chatbots useless. A calculator doesn’t “think” either, and it’s indispensable. It just means you should treat a chatbot’s confident explanation the way you’d treat directions from a confident stranger: appreciate the help, verify before you drive.

This week’s articles

Sixteen posts went live this week, our busiest yet. Here’s each one, in one line:

Like this? Get it every Saturday

Join the Plain Manual newsletter, one email every Saturday with the week’s articles and the news that matters, in plain English. No spam, unsubscribe anytime.

Sources we used this week

Related Posts

  • plain manual weekly 2 featuredPlain Manual Weekly #2: OpenAI’s Safety Exodus, Congress Goes After Rogue AI Agents, and 13 New PostsOctober 10, 2026Weekly Digestai news
  • plain manual weekly 3 featuredPlain Manual Weekly #3: AI Agents That Cheat in Testing, an 816-Billion-Parameter Model for Your Mac, and 16 New PostsOctober 17, 2026Weekly Digestai news

Leave a Comment