ChatGPT can now do your work for you

Happy Sunday,

Big week for ChatGPT. OpenAI just launched ChatGPT Work, an agent that takes a whole project off your plate and hands you back the finished thing, and it's the biggest release they've done in months. I'm breaking down exactly what it does, clearing up all the confusion around which mode you should actually be using inside ChatGPT and Claude, and ending with their new voice mode that finally makes talking to AI feel like a real conversation... and one very funny thing it still cannot do.

Let's get into it.

🚀 THIS CHANGES EVERYTHING

ChatGPT Just Launched an Agent That Does Your Work

OpenAI just bounced back big time. They launched ChatGPT Work, a brand new agent powered by their new GPT-5.6 model, and it's the biggest thing they've shipped in months. It's built for the big messy projects that take real planning and multiple steps. Not answering a question. Actually doing the job.

So here's how it works. You give Work a goal, and it gathers context from the apps and files you approve, writes up a plan, and then carries that plan out step by step, checking in with you on the big decisions along the way. It can stay on a single project for hours and hand you back the finished thing. A real spreadsheet. A real deck. A real document. Even a working website.

Now my honest take. This is OpenAI coming straight for Claude Cowork and they are not being subtle about it. But the thing that genuinely surprised me is the design. Making things actually look good has been the hardest problem for OpenAI to crack for years, and they've pretty much solved it. Look at the website it built me.

That came from a two sentence prompt. And if I kept tweaking it I guarantee you I'd end up with a genuinely perfect site. Their plugins and app connections are also significantly better than Claude's right now, it plugs straight into over 1,400 apps you probably already use.

And here's my favourite part of this whole story. Thirty-one minutes after OpenAI's announcement went live, Anthropic's Claude account posted a single sentence. No blog post, no explanation, nothing: "We've reset 5-hour and weekly rate limits for all users." They just quietly handed everyone their rate limits back in the middle of OpenAI's big day. It did five million views in half a day. Honestly? That feels a lot like Anthropic telling us they're a little scared. The rivalry has never been more real and it is so fun to watch.

You'll see this new mode right now in your browser and in the ChatGPT app on your phone. And if you use the ChatGPT desktop app, you'll want to grab the new one, which is actually the Codex app updated. Start throwing it the tasks Claude couldn't quite handle for you and see what comes back.

I also put together a full guide on exactly how to use Work to get the most out of your day to day. It's very in depth and covers everything you need to know to actually get value out of this thing.

Get my full ChatGPT Work guide →

Watch my full breakdown on Instagram → (and thanks for the sponsor, ChatGPT)

💡 SAVE THIS

Which Modes Should I Use in ChatGPT and Claude?

So now that you know what Work is, let me clear something up, because I know this confuses a lot of people. Both desktop apps now have three different modes and nobody is quite sure which one they're supposed to open. ChatGPT has Chat, Work, and Codex. Claude has Chat, Cowork, and Code.

And this matters way more than people realise. Almost everyone only ever opens Chat, gets a mediocre answer back, and quietly decides AI is a bit overrated. But using only Chat is like hiring the smartest employee you've ever had and then only ever asking them questions, never actually giving them any work to do. They'd be great at answering your questions. But you'd be missing 99% of what they're actually capable of. The mode you pick changes what you get back more than the prompt you write.

So here's the simple breakdown so you know exactly where to go when you open each app. Chat is for thinking. Questions, ideas, explanations, anything you want to talk through. Work and Cowork are for doing. This is where you hand over an actual job and walk away. Code and Codex are for building. Anything that ends up as working software.

And if you want each one laid out task by task, here it is.

And now the question I get more than any other. Which one should I pay for? My answer never changes. If you have the means, pay for both. I know that's not what people want to hear, but you can only get access to the best versions of these models when you actually pay, and they're genuinely good at different things. Being good at AI was never about picking a winner. It's about using the right tool for the right task.

Get my full guide to Claude's modes →

🎤 TRY THIS

ChatGPT's New Voice Mode Is Actually a Conversation

Told you it was a big week. OpenAI also just replaced ChatGPT's voice mode with something called GPT-Live, and it fixes the one thing that always made talking to AI feel awkward.

Until now, voice mode worked like a walkie-talkie. You talk, then it talks, and you take turns. The new one can listen and speak at the same time. So you can cut it off mid-sentence, jump in, change your mind halfway through, say "no wait, actually..." and it just rolls with it. It genuinely feels like a phone call now instead of leaving voice notes.

Then I came across a video online that made me laugh out loud. Someone asked this brand new, supposedly genius voice mode a very simple question. "How many E's are in seventeen?" It confidently answered two. So obviously I had to try it myself, and when I asked it the exact same thing it told me three.

What I found fascinating is that if you type that same question into the normal ChatGPT chat, it gets it right instantly. So why does the voice version fumble it? Because it's not the same brain. Voice mode runs on a completely separate model built for one job, answering fast enough that the conversation never feels awkward. We're talking a couple hundred milliseconds. Actually stopping to think a problem through takes whole seconds, and that would ruin the entire illusion, so it just fires back an answer on instinct.

So here's why this happens. These models don't read letters. They read words in chunks. A word like seventeen isn't nine separate letters to an AI, it's one or two blocks. So counting the letters inside it isn't something it can just glance at, it has to deliberately slow down and pull the word apart. The text version will happily do that. The voice version won't, because it's sprinting to keep up with you. Once you know that, a lot of weird AI behaviour suddenly makes sense.

So I went and asked Claude, Gemini, and Grok's voice modes the exact same question. Guess which one was the only single model to get it right. Gemini. I'm not going to pretend this is a serious benchmark of intelligence, but I did think it was pretty funny.

Counting aside, the thing that genuinely blew me away is live translation. Because it can hear and talk at the same time, it can sit between two people speaking different languages and translate back and forth in real time, with nobody waiting. If you travel at all, that alone is worth opening the app for.

It's rolling out on iPhone, Android, and web right now. Over 150 million people already talk to ChatGPT instead of typing to it, and after a few days with this version I completely get why. Open the app, hit the voice icon, and just talk to it like a person.

Go download the new ChatGPT desktop app and give Work a real project, not a test one. Something you've actually been putting off. That's the only way you'll get why this is such a big deal.

See you Wednesday.

Maverick

P.S. Want your business to actually show up inside AI? The brands that rank are seeing up to 3x the conversions. Get your free AI visibility score and see exactly where you stand inside AI.

P.P.S. If you're new here or you ever skip one, I've got a newsletter archive of all my past emails you can catch up on anytime.