47:51Riley Brown · Interview
OpenAI Merges ChatGPT and Codex
Riley Brown and Ras Mic dig into GPT-5.6, Codex's background computer-use, and why self-scoring agent loops are turning coding tools into a general operating system.
July 12thThe argument in one line.
Every major lab now ships a personal always-on agent, and the real shift this week is that these agents move into your phone, your messages and your company's org chart.
Read if. Skip if.
READ IF YOU ARE…
- You use ChatGPT, Codex or Claude daily and want one 17-minute briefing on what changed this week instead of scrolling X.
- You are deciding which agent platform (OpenAI Dot, Claude Duet, Grokbot) to build your daily workflow around.
- You work in a role like software, finance, HR, support or writing and want to understand how agent adoption is changing hiring.
SKIP IF…
- You want a deep tutorial on any single product; every update here gets a couple of minutes at most.
- You want unbiased benchmark analysis; the host states plainly which models he prefers.
The full version, fast.
The week's biggest agent news came from every major lab at once. OpenAI let users text their Dot agent over iMessage, route routines to specific channels, and spin up local Codex sessions by voice, then doubled Pro tier limits and restored the 20x Plus plan. Apple's Siri beta now has a built-in AI model, on-screen awareness and app intents. Google released Gemini 4 Carbon, which the host doubts after the underwhelming Argon. Anthropic is widening Claude Duet and filed an S-1 showing 900% revenue retention. Microsoft offered 60,000 employees voluntary exits, signaling agent-driven job shifts.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:29
01 · Intro
Previews the week: Dot texting, plan changes, Siri, Grokbot, Duet and the IPO.
00:29 – 03:28
02 · Text Your Dot from iMessage
Demo of setting up texting, routing routines to channels, and voice-launching a local Codex session.
03:28 – 04:41
03 · ChatGPT Plan and Usage Limit Changes
OpenAI doubles Pro limits and restores the 20x Plus plan.
04:41 – 07:56
04 · Siri AI Beta
Beta setup on iOS 27, the dedicated Siri app, app intents and on-screen awareness.
07:56 – 09:44
05 · Gemini 4 Carbon
Google's new biggest model, with skepticism after Argon and a stock drop.
09:44 – 10:36
06 · Grokbot Updates
Pages, a password vault, Apple location and calendar integration and better video.
10:36 – 12:44
07 · Anthropic Claude Duet
Duet rolls out more widely; the host wants local Claude Code access before switching from Projects.
12:44 – 14:04
08 · Anthropic S-1 and IPO
900% revenue retention and the argument that the best model drives growth.
14:04 – 16:59
09 · Microsoft Voluntary Exits and Agent Jobs
About 60,000 voluntary exits and the case for agent-implementation jobs.
Lines worth screenshotting.
- Texting your agent over iMessage means every connector you set up on desktop is reachable from your phone without opening another app.
- Agent routines can now target a specific channel, so an urgent email alert can arrive by text while a daily recap stays in the app.
- A voice message sent to a messaging agent can spin up a local Codex session that edits and pushes code on your own machine.
- OpenAI reversed its plan downgrade: the 20x Plus tier is back, and the $100 plan now carries 10x where it used to carry 5x.
- Siri's beta is a dedicated app with conversation history, on-screen awareness and app intents, but it still only handles simple actions.
- Strong benchmark cards did not stop Gemini 4 Argon from feeling worse than Claude Opus 5.5 to people who use AI seriously every day.
- Google's stock reportedly fell about 13% around the Carbon launch, which shows investors are not buying benchmark claims alone.
- Every major lab now has a personal agent platform: OpenAI Dot, SpaceX Grokbot, Meta Muse and Anthropic Claude Duet.
- Local coding-agent access is a non-negotiable for the host, and he says he will move from Claude Projects to Duet the day Duet can spin up local Claude Code sessions.
- Anthropic's S-1 reports 900% revenue retention: a customer who spent $1 a year ago is now spending $9.
- The host argues the best model has driven nearly all of Anthropic's growth, and Claude Code stickiness compounds it.
- Microsoft offered about 60,000 employees, roughly 27% of the company, a voluntary exit package on October 26.
- The roles most affected are software engineering, finance, HR and recruiting, operations, customer support and documentation writers.
- The host predicts agent teams will be larger than expected, because someone has to design, supervise and improve each agentic workflow.
Agents are leaving the app and entering everything.
WHAT TO LEARN
This week showed that personal agents now follow you into iMessage, Siri and payroll decisions, so the useful skill is managing agents rather than just chatting with models.
- Connect your agent to messaging early: one setup lets every desktop connector work from your phone, and routines can pick which channel they notify you on.
- Judge a model by daily use, not its model card; Gemini 4 Argon scored well but felt weaker than Opus 5.5 to people who use AI heavily.
- Local coding sessions are the feature that decides agent-platform loyalty, so check whether a platform can launch them before you commit your workflow.
- OpenAI's plan reversal shows usage limits can change fast; compare tiers by current multiples of usage rather than by name.
- Anthropic's 900% revenue retention suggests customers expand spending once an agent product becomes part of their daily work.
- Microsoft's voluntary exit targeted repetitive knowledge work, so roles built on summarizing, documenting and translating are the most exposed.
- The opportunity sits on the other side: companies need people who can map a department's tasks to agents and keep those workflows reliable.
Terms worth knowing.
- Dot
- OpenAI's personal agent inside ChatGPT. You name it, give it connectors and routines, and now you can text it from iMessage.
- Codex
- OpenAI's coding agent. It can run local sessions on your own machine to edit code and push changes.
- Claude Duet
- Anthropic's agent platform, an always-on agent organized into channels rather than separate agents.
- Grokbot
- SpaceX's agent platform, with pages, a built-in password vault, Apple location and calendar integration, and video generation.
- App intents
- An Apple framework that lets developers expose app actions so Siri can trigger them.
- S-1
- The registration document a company files with the SEC before going public, containing its financials and risks.
- Revenue retention
- How much revenue an existing customer group generates over time compared with a year earlier. 900% means nine times the spend.
Things they pointed at.
09:30channelMatt Palmer
12:32channelClaude Duet video by Riley Brown
Lines you could clip.
14:00
“The revenue retention is 900%. So if a customer spends $1 a year earlier, the same customer is spending $9 now.”
16:40
“In aggregate, there will be a job loss, but I think there's a ton of opportunity for people who want to go into these companies and implement agents.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
OpenAI made two huge updates over the past few days. Not only can you text your dot, which allows you to fully control codecs from iMessage, OpenAI also made huge changes to your ChatGPT subscription, which everyone is talking about. But way more happened this week in the world of AI agents, including Siri's biggest update ever for most iPhone users, Grokbot got a ton of new updates, and Anthropic is widening access to their new Duet platform, and they released some information about their IPO.
So the first weekly agent update this week is that you can now text your dot. So now on ChatGPT, if you navigate in the top to your dot, you'll see whatever you named your dot, which is your own personal agent. Now they added a new feature.
If you click directly on your dots icon, let's go ahead and do it right now. You can see here it says set up texting. So I'm going to hit set up texting.
And now it says set up texting. Get started. We're going to click get started.
I'm going to put in my phone number secretly. And now it sent me this six digit code. I'm going to click on it, hit verify.
And now I can press open messages. And here it auto loads a message called Hey Dot. And I just sent a message to my dot and it just read the message.
And there you go. Hey, Riley, same dot now in your text. Send me a video idea or a link when you're out and we can work through the hook right here.
So it has my context already. What plugins or connectors have we set up that you have access to right here? And now you can see that my dot is typing.
Now it left this. Little eyeballs here. And there you go.
And so all of the connections that I've set up inside my GPT desktop app, I can use directly from iMessage, which is awesome. And what's cool about DOT is you can create routines or automations. You can say every morning at 9 a .m., I want you to check my email and give me a summary.
Every day at 2 p .m., I want you to check AI News and send it to me. Well, now you can distinguish between the... channels it messages you through.
So I can say something like this. Hey, every day at 530, I want you to do like a little recap for me. And I want you to send me a message on text message.
I want that to be over text. So I can send this to my dot. And it just said, I'll set up a short daily recap by text at 530 p .m.
And I'm just as a test. Hey, can you please just say hello? This is Dot from the computer.
Send that to my iMessage so I know it works. I can kind of send things to different channels.
And so let's see. I just got this message. Hello, this is Dot from the computer.
So now you have Dot in different channels, which is super useful because sometimes you're only on your phone. You don't want to get a message over the ChatGBT app. Maybe you have a really urgent email thread and you could just tell your Dot, hey, if something happens in this thread, I need to know about it.
Please text me. And then, of course, I can fully control Codex. And what I can do, look at this.
So I'm going to press this. Hey, buddy, I need you to please spin up a Codex session, which just. updates, make a slight change to my landing page on agent native and just kind of clean it up and just make some minor tweaks to clean up the design and then push your changes, please.
This should be done on a local codec session, please. So look at this. I can send voice messages to my dot and then it will spin up a codec session and handle the task.
You can do this for my message, which is really cool. But this isn't the only update from OpenAI that many people are talking about. There's also some huge new updates to the OpenAI or ChatGPT plan.
So here's the announcement from Thibaut. We are doubling all usage limits for all pro 100, 200 and 500 tier plans and business premium effective today. We heard your kind feedback and won't be reducing usage limits for pro 200.
So your work and codex usage will stay the same. And previously open AI had made some announcement that made people upset. So basically the 20 X plus became 10 X plus.
And then they basically reversed it now. So it's back to the original 20x plus. And you can pause the video right now to get an idea of like what each plan is at compared to what it was before September 29th.
But people seem to be much happier now with the plans. And in some cases, it's actually more than it was before. Like the $100 per month plan now has 10x plus where it originally had 5x the plus plan.
Long story short, more codex use. more chat GPT usage, and more dot usage. For the next update this week for the world of AI agents is really underrated and no one is talking about it.
Siri AI is in beta and you can download it right now. I'll show you how. Before I show you how, let me just show you how this works.
So I'm here using my iPhone. And if I hold the Siri button, hey, can you please search the internet and tell me a little bit about Riley Brown's latest videos? And it now actually has an AI model built in and we can expand it.
And now it's just like chatting with some AI model. They don't tell us what AI model. That's not really Apple's style.
They're just, it's Apple intelligence. Can you please summarize the things we've been talking about the past few days? Over the past few days, we have discussed several topics, including Riley Brown.
We talked about your background as an AI educator and content creator, your latest videos regarding OpenAI's Codex and the Bluey app, and your role at VivaCode. And now... It's not just talking to Siri.
There's actually a dedicated Siri app. So if you search for Siri on your phone, you can click Siri. And now we have recent conversations.
In order to set this up, you should go to settings on your iPhone. You're going to do is you're going to type in update and you're going to click on software update here. What you will do is you will click on beta updates.
It'll show you if you can update your Siri to the latest version, but you do need to be on iOS 27 developer beta, and then you need to enable the Siri beta. And so Apple is releasing something called app intense, or they already released it. And you can basically, when you're developing an app, you can make.
Actions accessible from Siri. And so I can do some actions on my phone. Just a quick example, just a simple one.
Hey, open the Clod app for me, please.
So it opened the Clod app. That's nothing crazy. Type in, please, based on your memory of me, what are my greatest strengths?
Okay, well, it's not. super insane right now you can just do very simple things with siri but it's starting to become more intelligent and with so much competition going on in the like personal agent space there's no one set up better than apple and so if they can get this right and you can just fully control your phone with siri and then hopefully your computer with siri It'll get really interesting.
And so I just wanted to put this on your radar because it is getting smarter. And so the next two things I want to talk about are personal context as well as on -screen awareness. And so if we go to our phone right here, it actually sees everything on my screen.
So I can use Siri. Hey, look at this on my screen. Can you please do a little research on this?
And look, so it has that on -screen awareness, which is really interesting. And then you can do a follow -up question. This is pretty useful.
This one's not crazy, but I actually found this to be useful yesterday. Hey, there's like an app. It's like love themed.
It allows you to build apps with AI. Can you open it?
And there you go. It opened the lovable app here. So we're just starting to see Siri get a little bit smarter, which is useful across all of your Apple products.
And I'll be covering this as it gets better. I hope it gets a lot better. OK, so now it's time to move to Gemini.
So yesterday they announced a new model called Gemini for Carbon. This comes two weeks after Gemini for Argon, which, in my opinion, was a huge disappointment despite having really good benchmarks. And so yesterday, Sundar said, I did say we'd be releasing our models periodically.
Introducing Gemini 4 Carbon, our next frontier model. Now, when you look at these model cards, you will see, oh my god, it is better than Claude Opus at every category, except for one. It is better than GPT -6 Astra at every category, except for one.
And remember, they released their model card for Gemini 4 Argon. And look, it looks very similar. By looking at this, you will see that it is better than Opus at all but two categories.
And so Gemini 4 Argon is the smaller of the model. Carbon is the one that's coming out soon that's bigger. And this one appears to be better.
I haven't met a single person who seriously uses AI who thinks that Gemini 4 Argon is better than Claude Opus 5 .5 at anything. Anything at all. And so that is why I am continuing to be skeptical about the up...
And they released this model on days that they released their earnings and, you know, they're down 13 % from before it dropped. So investors are not super bullish on this new model. either.
And so what you need to know is it is a new model. It is their biggest model yet, but their previous model was incredibly underwhelming. And I really genuinely hope this is good.
Google, if you're watching, I hope this model is great, but I'm not sold yet. Okay, so now I want to talk about Grokbot because they released some new updates. Now, I am not going to go super in -depth on the Grokbot platform because you know who does that?
Matt Palmer. I highly recommend following Matt Palmer to stay up to date on all of the Grokbot updates. But basically, as a quick rapid -fire overview, they release pages in Grokbot.
So this is similar to GPT Space or Cloud Artifacts. They also release their own password vault that is separate than 1Password, but you can add passwords directly into Grokbot and it can use them to sign in. things they have more apple integration so now you can ask rockbot about where you are because it's connected to your location on apple you can connect it to your native calendar app which is really cool grokbot can also make videos much better now they made some improvements there and those were the most notable in my opinion okay moving on to the world of anthropic so anthropic is rolling out duet This is their agent platform.
OpenAI has Dots, Grok or SpaceX has Grokbot, Meta has Muse, and Anthropic now has Claude Duet. And by the way, you see this presentation? I plan this out all directly inside Claude duet and it is honestly very fun to use.
Now I've been very vocal about how the best Claude feature is Claude projects and Claude projects. Ironically, I named my Claude project duet because it's a very similar app. You know, it's very similar, except in quad duet, I can spin up local quad code sessions, which is incredible.
It's like a non -negotiable for me with these agent platforms. Now, the same way that dots can spin up a local codex session on projects, I can spin up a local quad code session. And so I was a little upset vocally on Twitter about how I was disappointed that on quad duet, you cannot spin up a local quad code session.
Luckily, being vocal on Twitter has its benefits because I tweeted here about my annoyances with Claude duet. And this guy gave Molly from anthropic said, really appreciate the feedback.
We're working on remote control and local Mac access more to come on that soon. And they also said they only use Opus 5 .5. I said, it feels like Opus 5 .5 light.
So I think that this might just be an error in their system prompt for Claude duet, but I anticipate if they are using Opus 5 .5, this is going to get better over the next few weeks. And so if they add local Claude code access. I will go back and I will be leaving Claude projects and I will basically spend all my time in Claude duet because it is your agent that lives 24 seven and it will be able to use Claude code.
And one of the main reasons I like Claude Duet is they have these channels over here. So instead of having different agents, you just kind of have different channels. I have an entire video on Claude Duet, and that can be found right here.
I made it six days ago. So if you want a full video on Claude Duet, go ahead and watch that video. And then the final update that I wanted to include, because I think it has big implications for our future of using agents, and that is they have an IPO coming up and they released a new document called the S1, which is this giant report that you.
have to create before you go public with all of the company's data and everything. Oh, by the way, I created this on a claw duet from my phone, which is pretty cool. The most interesting slide to me, and like it went through all their revenue, their revenue numbers are just insane.
Their one thing that I want to talk about is their revenue retention. You know, the revenue retention is 900%. So if a customer spends $1 a year earlier, the same customer is spending $9 now.
And what came through to me is how sticky of a product this is. Their products have been since they released cloud code.
And I'm starting to realize that like the model. is just most of the importance. They have consistently had the best model over the past two years.
And this has been the cause of all of their growth, in my opinion. And they are just so locked in and focused on creating the best model. And it shows in the S1.
So I highly recommend looking at the S1 or getting AI to create a document. And I will link this in the description if you want to check out all of this information on the Claude S1, because they're about to go public. And you can think about investing if you want.
I'm not endorsing it as an investor. I don't do anything like that. And then finally, we're going to finish off with some negativity.
I don't like to make this channel about the negatives. I think there's plenty of people doing that. There's a lot of fear mongering going on, but I do want to flag this as an update in the world of agents because a lot of the work that companies are doing, especially the repetitive work starting to be replaced by agentic workflows.
And there seems to be the beginning of another round of mass layoffs. And I predict that like if the economy starts to contract a little bit and other companies like Microsoft are starting to lay off a ton of people, I think a bunch of companies could be laying off people, especially if they're in high risk jobs, which I'll get to in just a second.
But what happened was is Microsoft made an announcement and they basically gave 60 ,000 people, a voluntary exit on October 26th, which is about 27 % of Microsoft, which is truly insane, which means you can leave. You can take this package.
We'll pay you this really nice severance package. If you leave voluntarily, they're basically doing a pre layoff, which is very scary position to be in as an employee. And the most interesting part were the job type.
So a lot of software engineering jobs, a lot of finance jobs, HR and recruiting and operations. based jobs, customer support, and writers, people who are writing documentation, doing summarizing, translation, et cetera. These were the top jobs that are being offered this voluntary exit.
And so these are a lot of the things that AI is going to get really good at. And so Microsoft internally is getting ready to just implement agents all over their business. And so the people who are being replaced are, you know, there's going to be people that come into the company after these people leave, and those are going to be.
the people who truly understand how agents work and basically can look at an entire department within Microsoft and think, okay, how do we use agents to accomplish all the tasks that we can? And I think these teams are going to be bigger than people think. It's just kind of a shift from people who do manual work that include a lot of these things to a team of people who set up agents to get everything done.
And so, yes, in aggregate, there will be a job loss, but I think there's a ton of opportunity for people who want to go into these companies. implement agents and you need a lot of people to manage these agents, to make sure that they're doing the right thing, to add different agentic workflows, to improve the efficiency of things.
And so there's a lot of opportunity on the other side of a lot of people, unfortunately losing their job. And so I'll be covering this a lot more going forward, like what the actual opportunities are in the job market for implementing agents into companies. It's one of my biggest passions right now.
I'm working with some big companies right now on how to implement agents. And there's so much demand for people who truly understand these workflows. And I'll be talking about it a lot in the future.
So thank you guys so much for watching. This was an update. Please subscribe.
Please like. I love you forever. I'll see you here in the next one.
The Hook
The bait, then the rug-pull.
Riley Brown packs a full week of AI agent news into 17 minutes, built around a single grid slide that fills in as each update lands. The through-line: personal agents are moving into phones, messages and corporate payrolls.
How they asked for the click.
VERBAL ASK
16:50subscribe
“Please subscribe. Please like. I love you forever.”
Short and casual, placed after a forward-looking promise to cover agent job opportunities.
MENTIONED ON CAMERA
Visual structure at a glance.
