A week of AI-agent news: Codex takes over iMessage, GrokBot draws Claude-Code-moment comparisons, Anthropic ships /design and mobile Cowork, and Slack gives agents their own channels.
Posted
yesterday
Duration
Format
News Roundup
hype
Views
6.6K
215 likes
57 · 43
Big Idea
The argument in one line.
OpenAI's Codex can now read and send iMessages on approval, Slack shipped dedicated agent-native code channels, and the real fight between GrokBot, ChatGPT, and Claude comes down to ease of use versus model quality versus knowledge-work output, not raw agent capability.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You already juggle ChatGPT, Claude, and Grok or GrokBot day to day and want a fast, opinionated read on which one to reach for and why.
You're curious about giving an AI agent access to your actual iMessages or Slack and want to see what the approval flow looks like before trying it yourself.
You run a small team and want a read on Slack's new agent-native code channels before your team starts using them.
SKIP IF…
You're looking for a hands-on tutorial with exact setup steps — this is a demo-and-reaction news roundup, not a walkthrough.
You don't use ChatGPT, Claude, or Grok in a work context — none of the platform comparisons will be actionable for you.
TL;DR
The full version, fast.
This week's biggest AI-agent news: OpenAI's Codex can now read and send iMessages through an approval-gated workflow, including retroactively analyzing a 3,000-plus-message group thread. GrokBot, Cursor's agent super app, is drawing early converts despite a weaker underlying model, so Riley scores GrokBot, ChatGPT, and Claude live across five categories and lands on a split decision: GrokBot wins ease of use, Codex wins the in-app browser, and Claude wins knowledge-work documents. Anthropic separately shipped a /design mode inside Claude Code, moved Cowork to mobile and web, and added terminal output styles. Slack launched agent-native code channels with Anthropic, GitHub, Cognition, and Vercel agents built in.
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.
Riley previews the episode: Codex controlling iMessages, GrokBot's rise, Anthropic's Claude Design and Cowork updates, and the new Slack Code agents.
00:51 – 04:31
02 · Codex takes over iMessage
Codex's new @messages feature drafts and sends iMessages through an approval flow, then analyzes a 3,251-message group thread for communication feedback.
04:32 – 05:23
03 · Sponsor: HubSpot's AI agent cheat sheet
Riley reads the HubSpot-sponsored plug for a free cheat sheet covering seven agent tools.
05:24 – 07:28
04 · Why GrokBot is suddenly everywhere
SpaceX's Cursor-owned super app GrokBot is winning fans for its remote computer and one-click setup, but Riley and an insider agree the underlying Grok 4.6 model is the weak link until Grok 4.7 lands.
07:29 – 09:46
05 · Building the scorecard: usage and ease of use
Riley draws a live usage split (GrokBot 35%, ChatGPT 50%, Claude 15%) and a scoring table across five categories, awarding GrokBot the ease-of-use point for its simple, task-scoped bots and routines.
09:47 – 10:59
06 · Browser and model points go to Codex and Claude
Codex's built-in side browser wins the in-app-browser category outright; on model quality, Claude and GPT 5.6 Sol are judged ahead of Grok, with Claude specifically credited for knowledge-work documents.
11:00 – 13:12
07 · iOS showdown: real-time voice spins up a Codex session
GrokBot's iOS app is simplest to set up, but Riley gives ChatGPT the win for a real-time voice mode that can spin up a full Codex coding session mid-conversation — he demos it live building a landing page.
13:13 – 14:36
08 · The landscape: nobody wins outright
Riley closes the comparison: GrokBot wins on ease of use, Codex on the browser, Claude and GPT are neck-and-neck on models, and Claude still wins on finished documents.
14:37 – 16:48
09 · Claude Code gets a /design mode
Typing /design inside Claude Code's desktop app invokes Claude Design as a skill, drafting UI options on a side canvas that Claude only implements once you pick one.
16:49 – 18:37
10 · Claude Cowork lands on mobile and web
Cowork, Anthropic's cloud-agent mode, is now available on phone and web instead of desktop-only, though Riley flags that the line between a normal Claude chat and a Cowork task is still confusing on iOS.
18:38 – 19:15
11 · Terminal output styles: pick how Claude talks to you
A new /config setting in the Claude Code terminal lets you choose a communication style — proactive, explanatory, or learning.
19:16 – 22:13
12 · Slack ships agent-native Code channels
Slack Code launches with Anthropic, GitHub, Cognition, and Vercel agents that spin up dedicated code channels so a whole team, not just one developer, can steer an agent together on a build.
22:14 – 22:48
13 · Wrap-up: the shift to multiplayer agents
Riley recaps the episode and argues the second half of 2026 is about moving from personal agents to teams of agents and eventually multiplayer human-agent collaboration.
Atomic Insights
Lines worth screenshotting.
OpenAI's Codex can now read, draft, and send iMessages from a Mac, with every outgoing message gated behind an explicit approval step.
The same Codex messaging feature can retroactively analyze a 3,251-message group thread and produce a communication-style critique in under a minute.
GrokBot ships with its own remote computer, removing the need to buy a dedicated Mac mini just to keep an agent running around the clock.
Multiple early GrokBot users report replacing their entire personal-agent stack with it, despite calling its underlying Grok 4.6 model weak at writing and knowledge work.
ChatGPT's real-time voice mode can spin up a full Codex coding session mid-conversation, turning a spoken request into a running build task with no keyboard involved.
In a live three-way scorecard, GrokBot won ease-of-use and simplest iOS setup, Codex won the in-app browser, and Claude won knowledge-work document quality — no single platform swept every category.
Claude Code's new /design command invokes Claude Design as a skill inside a coding session, drafting UI options on a separate canvas that only get implemented once you explicitly approve them.
Claude Cowork, Anthropic's cloud-agent mode for long-running tasks, is now available on mobile and web, not just the desktop app.
Claude Code's /config now offers a choice of communication styles for the terminal — proactive, explanatory, or learning — controlling how much Claude explains alongside the code it writes.
Slack Code spins up a dedicated code channel per build task, so a whole team can see and steer the prompts sent to a coding agent instead of one developer working in isolation.
Anthropic, GitHub, Cognition, and Vercel all shipped agents into Slack's new agent-native surface at the same time, suggesting 'agent lives in team chat' is a shared industry bet, not one company's feature.
The video frames 2026 as a two-part shift: the first half was about personal agents, and the second half is about teams of agents and eventually multiplayer human-agent collaboration.
Takeaway
Five updates, one theme: agents are getting write access to your real accounts.
WHAT TO LEARN
Codex can now send your texts, Slack can now host a whole team's coding agent in its own channel, and the platform fight between GrokBot, ChatGPT, and Claude is decided by task fit, not one winner.
02Codex takes over iMessage
OpenAI's Codex can now read, draft, and send iMessages from a Mac, with every outgoing message gated behind an explicit approval step.
The same Codex messaging feature can retroactively analyze a 3,000-plus-message group thread and produce a communication-style critique in under a minute.
Treat an AI-generated group-chat analysis as a mirror, not a verdict: it can surface a real pattern worth acting on even from an imperfect tool.
04Why GrokBot is suddenly everywhere
GrokBot ships with its own remote computer, removing the need to buy dedicated hardware just to keep an agent running around the clock.
Early GrokBot users report replacing their entire agent stack with it, but the consistent complaint is the underlying Grok 4.6 model, not the product design.
Separate the container/product experience from the model quality when evaluating an agent platform — they can lag or lead independently, and a promised model upgrade can flip the calculus in weeks.
05Building the scorecard: usage and ease of use
GrokBot's ease-of-use edge comes from scoping each agent to one narrow task rather than one general-purpose chat you have to steer yourself.
An iOS app that auto-connects to your desktop setup with zero remote-configuration step removes a real onboarding barrier most agent tools still have.
Track your own real usage split, not stated preference, when comparing agent platforms — it's the more honest signal of what actually earns a place in your workflow.
06Browser and model points go to Codex and Claude
An agent with a built-in in-app browser can open and act on a live webpage or document without leaving the chat, a meaningfully different experience than one that can only describe what it would do.
For knowledge-work output like spreadsheets, presentations, and front-end design, model quality still separates the platforms more than interface polish does.
07iOS showdown: real-time voice spins up a Codex session
ChatGPT's real-time voice mode can spin up a full Codex coding session mid-conversation, turning a spoken request into a running build task with no keyboard involved.
The distinction between a platform's ambient voice assistant and its remote voice-to-agent mode matters: only the latter can trigger an actual coding session in the background.
08The landscape: nobody wins outright
No single agent platform wins across ease of use, model quality, and output polish at once; the practical move is picking per task rather than picking one platform for everything.
A platform can be someone's most-used tool for reasons that have nothing to do with raw model quality — onboarding ease and task-scoped agents can outweigh a weaker model.
09Claude Code gets a /design mode
Claude Code's /design command invokes Claude Design as a skill inside a coding session, so it can propose UI options without touching your actual running app's code.
The design draft happens on a separate canvas; you must explicitly tell Claude to implement the version you picked, keeping AI-generated design suggestions from silently landing in production.
10Claude Cowork lands on mobile and web
Claude Cowork is now available on mobile and web, not just desktop, closing a gap with ChatGPT's chat/work split.
The UI distinction between a normal Claude chat and a Cowork task is still confusing on iOS specifically, since there's no separate tab like on the ChatGPT app.
11Terminal output styles: pick how Claude talks to you
Claude Code's /config now lets you pick a communication style for the terminal, controlling how much explanation you get alongside code changes.
This is a workflow-fit setting, not just a preference toggle: a learning-oriented style suits someone trying to understand a codebase, while a proactive style suits someone who just wants the change made.
12Slack ships agent-native Code channels
Slack Code spins up a dedicated code channel per task instead of running an agent inside your regular team channels, keeping the build conversation separate from day-to-day chat noise.
The whole team, not just one developer, can see and steer the prompts going to the agent in that channel, a real shift from the solo 'agent on my machine' workflow most coding agents still assume.
Anthropic, GitHub, Cognition, and Vercel all shipped agents into this Slack surface at once, signaling that agents living inside team chat, not just an IDE, is becoming a shared bet across vendors.
Glossary
Terms worth knowing.
Codex
OpenAI's coding agent product, extended in this update to control macOS iMessages through an approval-gated workflow.
ChatGPT Work
The 'Work' mode inside the ChatGPT desktop and web app for longer, more complex agent tasks, kept separate from a normal chat.
GrokBot
A standalone AI agent 'super app' from Cursor's owner that runs narrow, task-scoped agent 'bots' and comes with its own remote computer.
Claude Cowork
Anthropic's cloud-based agent mode for long-running tasks, now available on mobile and web in addition to the desktop app.
Claude Design
Anthropic's separate design product, reachable from inside Claude Code via the /design command as a drafting skill.
Slack Code
Slack's new agent-native surface that spins up dedicated 'code channels' where a team and a coding agent work together on a build.
Output style (Claude Code)
A /config setting that changes how much Claude explains its actions in the terminal, with options like proactive, explanatory, or learning.
Fable
The name used in this episode for Anthropic's newest top-tier model, referenced when comparing model quality across agent platforms.
“First half of 2026 was about the personal agent, and I think the second half of this year is all about the transition into creating a team of agents.”
the episode's closing macro thesis, quotable on its own→ IG reel cold open↗ Tweet quote
The Script
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.
17px
metaphoranalogy
Huge updates in the world of AI agents. OpenAI just released a brand new update where you can fully control your iMessages inside Codex and ChatGPT. And I'll show you exactly how to set this up and how to use AI to control your iMessages.
But we have way more to talk about today as well, like the latest Grockbot updates and their rising popularity. We also need to compare the different super app platforms like Grockbot, Cowork, and GBT Work.
And we also need to talk about how Anthropic is merging Claude Design and Claude Code. Also, how Anthropic just released Cowork on their iOS app along with a course to learn all of the different tools. And then finally, we need to talk about the brand new Slack code, which was released yesterday, which allows you to add Codex and Quad directly into Slack.
This is agent native. I'm Riley Brown. I do these updates every week.
So for the first agent update this week, we have a brand new feature inside Codex that was just released a few hours ago. If I type at messages, you can see this looks like Apple messages.
Please, can you send Emily a message saying, hi.
Did you get the groceries? And here it says, I'm using the messages texting workflow to find Emily and send the exact text and verify the result.
And look at this. It has this little approval thing. Send this to Emily Lambert.
Hi. Did you get the groceries? And it has a dash codex.
I can remove that, and I can edit the message, and I can hit continue. And there you go. It's sent to Emily.
Let's see if it's sent. I can pull in my phone, and you can see here it's sent, hi. Did you get the groceries?
Send another one which says, I'm really hungry.
So I'm just showing you how this works. We're gonna do a much more practical use case for businesses in just a second. It does require for you to have the ask approval on.
It won't just send without asking for approval, but here we go. We can hit continue. And as you can see, it popped up here in the iMessage.
Here, I was using it inside Codex, which I use for most things when I'm using the ChatGPT desktop app, but we can always switch to ChatGPT mode. And then if we create a new chat, we can select work. And then I could say something, can you please look through my conversation with Anj and Vishal?
We have a group message together, and I want you to analyze it deeply, basically, all of the things that we've talked about. And I want you to make one suggestion to me in general on how I can improve my communication style and my ability to work with others as a team.
And I want you to suggest one message to them based on that and the next thing that I should say to them.
And so you can have AI. If you have a super long group message, maybe it's with your team, is you can have AI fully analyze all of those conversations.
It doesn't matter how long they've gone on for. And notice here that I didn't at mention messages, so I can just go at mention messages and make sure that it gets it.
And you can see here it found the exact group thread. I mean, look at this. It says the thread is substantial.
3,251 messages across two versions of this group from May 14 through today. Okay.
So it just analyzed 3,251 messages, and this is pretty insane.
And it's probably true. It says your strongest communication trait is your ability to create energy and and quickly turn messy ideas into sharp positioning. You're also generous with encouragement, willing to help, emotionally honest, and capable blah blah blah blah blah.
The one thing I would change is make it unmistakable whether you are brainstorming, asking for input, or making a decision. You often communicate all three with the same intensity.
A passing idea arrives as we need to do this ASAP. And I can't really show all of this, but this is a really good way of, like, analyzing how you act in a group.
And I find this to be a pretty pretty brutal yet useful practice to do with AI with this new messaging feature that was just added to Codecs.
And then it gave this little feedback right here. I'm gonna say, okay. Send it.
And then we can fully approve it. And now it's sent. And as you can see here, it's sent on my actual iMessage.
And I even sent them a screenshot of this feedback, and they're like, damn. Useful. And, uh, there's the message from Kodak.
If you want an agent actually working for you, then this free AI agents cheat sheet from HubSpot breaks down everything you need. It breaks down seven tools, including Claude Code, ChatGPT, agent mode, Zapier, and n eight n.
For each one, you get what it's best at, how long setup takes, what it costs, when to use it, and when to skip it. And they include copy and paste starter prompts for real workflows, competitor research, organizing files, routing leads, and my favorite is the Manus prompt.
You drop in a video link, and it watches it and writes your full YouTube description with time stamps. That one alone is worth the download. So if you want a simple starting point, grab HubSpot's free AI agent cheat sheet, and thanks to HubSpot for sponsoring this video.
So the next update that I want to discuss is the rise in popularity of GrockBot, which is Cursor's super app.
And remember, SpaceX acquired Cursor, so it's actually SpaceX's super app. And many people are saying that they think bot is another Claude code moment for AI.
I I think it's a little too early to say that. Um, BridgeMind said, I've replaced all of my Hermes agents with GrockBot. GrockBot comes out of the box with its own remote computer, which makes the setup way easier, and it also removes the need to purchase any other Mac minis.
Uh, Lenny said he got early access to GrokBot, and he's hooked. Preston said GrokBot is proof that certain ideas haven't really been tried until they're executed perfectly. And then we also have news from someone at SpaceX who said the following.
And I want you to keep in mind, in my opinion, the thing holding Grock Bot back straight up is the model. The user interface of Grock Bot is the best in the world, in my opinion. I think how easy it is to set up the iOS app is best in the world.
I will say Grock 4.6 is just not that great of a model. It's not good at writing. It's not really good at any knowledge work tasks.
If you ask it to create a document, it doesn't do that good of a job compared to GPT 5.6 Soul or an even the Anthropic Fable models. It's just not quite there in terms of knowledge work. However, now from SpaceX said Grok 4.7 will be better at writing in bot, uh, which is Grokbot and in general.
I'm very excited for that. For now, my tip is to feed your Grokbot examples of your previous handwritten content. It will learn over time and adjust to your style.
Grok bot remembers and will get better with usage, which is encouraging. So Grok 4.7 apparently is coming out in about two to three weeks, and I heard it's going to be better.
I do think this platform, if it had a model as good as 5.6 Soul, I genuinely believe it would be, like, my favorite tool to use for all agent work. And so I wanna take some time real quick and talk about the differences between these different platforms and where I think each one has an edge.
And I think, first of all, this is about my usage in terms of how much I use each platform, and that is Grokbot, Codex slash ChatGPT, and then Claude, you know, whether it's their iOS app or their desktop app.
This ends up being the amount I use each one. And so the first thing that I'm going to do is I'm going to say that the ease of use is won by Grokbot. There's something incredibly easy to understand about you have your little bots, You can create a new bot, or I can just immediately go to my sponsorship bot or my content bot, and I can just say, please check the videos and Slack.
And this is videos in at Notion, um, and Slack. Please tell me what I need to do next.
And there's something really nice about having a little bot for a specific task.
Right? This to do list bot, all of this does is it literally just tells me exactly what I need to do, and it just, like, reminds me every day of all the things that I need to do. And then on top of this, their iOS app was just incredibly easy to set up, and it automatically connects with the desktop app as well.
Like, you don't need to set up any remote, which I thought was interesting. And then the final thing that they did to make this really easy to understand is that each agent has their own routines.
For instance, on the iOS app, we can very easily just go to the content agent. If we click on the content agent, we can see the routines that this agent has.
And if you compare this with the ChatGPT app, it's hard to tell, like, where you go if you wanna do a work task. Do you go to codex? Um, what's the difference between ChatGPT and work?
I genuinely believe that a lot of people are confused about this, which is why I made an hour long video because GPT work is incredibly powerful. It has a really good in app browser that I really like that Grockbot does not have. It's just a little bit more confusing than Grockbot, which is a standalone product that is pretty easy to understand.
And so when it comes to ease of use, definitely Grokbot gets the point. When it comes to the in app browser, no one comes close to Codex. I still think it's the most underrated part of using the desktop app is that it has a built in browser.
And the agent, when you ask it to go to a website or create any document on Notion, it opens it up in this side browser. And so, honestly, I think the dream user interface for a super app would be like agents on the left, just like GrokBot right here, except it could open up a browser on the right.
I think that would be absolutely amazing. And then when it comes to the models, I personally believe that unless you are a massive company with a huge budget to spend on Fable, I think ChatGPT has the slight edge in the model because GPT 5.6 Soul is much better than Opus, Claude Opus five, which I actually don't really think is that good.
So if you can afford, uh, Claude, then the model is best with Claude.
But for I would say that on average, it's probably a toss-up. I will say this is the biggest downfall. We're gonna give them a big giant red circle here, Grock Bot.
Grock Bot gets a giant red circle because this is a negative three points. Their model is nowhere close to ChatGPT and Claude's best models, and that is one of the reasons why I use Grokbot less than ChatGPT codecs.
Their model's just not quite as good. And I will say, in terms of Anthropic, the next point that I'll give them is knowledge work and docs. So these are things like spreadsheets.
So for spreadsheets, and then also, like, presentations.
And then this also includes, like, front end design. Like, Anthropix models are just better, and that gives them a huge edge.
I just think they have a clear edge here. In terms of iOS app, it depends on what you want. If you want the simplest iOS app, I would go with GrockBot.
It's incredibly easy to set up. It's just like a little team of agents, so I'll give them a half point. But if you wanted the most powerful, I would say the ChatGPT app with the Codex remote, the fact that you can connect to Codex remotely and you can also use real time voice, I think that's what gives them the dub here is the fact that this app has real time voice and it's really good.
And I'm not talking about the normal real time voice here on the ChatGPT app. So if you navigate to ChatGPT, they have a real time voice version here, which is pretty good.
And you can connect it it connects to your integrations and plugins and stuff, which is awesome. But what I'm talking about is in the remote section, the real time voice this version of real time voice can actually spin up codec sections.
Isn't that, uh, sessions? Isn't that right, codecs? I'm ChatGPT.
And, yeah, I can help with that kind of work, talking through IT. So you could spin like, if I asked you to spin up if I asked you to spin up a Codex chat right now that, uh, builds a landing page which compares GrockBot, ChatGPT, and Claude, you could do that right now. Can you please do that?
Sure. Let me put that together. And what's really cool is it will just spin up a new chat.
If we were to navigate to chat GPT, you can see that this latest voice conversation is showing up here.
Everything that we said would show up right here.
And you can see here it just created a brand new codec session right here, and it says build a polished and responsive one page landing page that compares GrockBot, ChatGPT, and Claude. So that's really cool.
You can use real time voice to spin up codec sessions. And so for that reason, I believe that codecs probably gets the point here.
And so this is kind of the landscape. Grokbot is winning because it makes agents very easy to understand and easy to use and easy to get started, whether it's on desktop or iOS. ChatGPT and Claude are neck and neck in models, but it's the worst advantage for GrokBot.
And then the iOS app, I believe, is just the strongest, and the browser's the strongest with ChatGPT. Whenever I create a Notion doc, Google Docs, everything, I just have it open up directly in the Codex browser, and I'm signed in to everything on the Codex browser. But the main reason I still use Claude is that it just creates better knowledge work documents, period.
All of the types of documents that it creates are better. Even Opus and Fable, especially Fable, does an incredible job at creating, like, really high quality PDFs. So for those reasons, I use Claude.
And so that's kinda my take on the landscape of these different super apps. So speaking of Anthropic and Claude, let's move to our next agent update, which are a series of updates from Anthropic. So the first update from Anthropic is that directly inside Claude code, specifically using it in the desktop app, meaning within the desktop app on Claude.
So if I'm running Claude here and I'm using the Claude code app, I can use slash design.
And this will invoke design mode, which is a separate mode. Right?
If you go home and you click on design mode, this is a whole separate product for Claude, and it's really popular.
It has a huge audience in and of itself, and it is really good. They're they're adding some of these features from Claude design directly into Claude code, which I think is really cool. But here's a video.
So you can type in slash design, and it says redesign the composer based on what functions people use most. And because it's connected to Claude Code, you can get it to use Claude Code, and then it just uses Claude Design basically as a skill.
And so Claude will actually draft it here on the side. And so this isn't actually adding your app or changing the code in your app. It'll look at your app, create a little draft, and then you can choose which one you want to implement in your actual app, which is really cool.
And you can fully customize it.
And and that is the most important part. I made a new artboard with some riffs. Implement that.
So it allows you to, like, type in your design ideas. You can alter them on in this little, like, mini Figma over here when it uses the design skill. And then it said, I just made a new artboard with some riffs.
Please implement that design that we created in the actual app that you're building. And so it just gives you a lot more control over the design. Huge fan of this update.
And then a few days ago, Anthropic made Claude Cowork available on mobile and web for all paid plans. And so this is Claude Cowork, which is much more similar to ChatGPT work because it's an agent running in the cloud.
Claude Cowork used to only be for the desktop app. You couldn't access the cloud version of Claude Cowork from your phone, and now it's available on phone and web. And on web, it's accessible by toggling on this Cowork tab right here.
This is similar to the OpenAI paradigm where you have chat and co work. On the desktop app, it's in the same exact place. You have chat and co work, and this is on the home tab.
But on iOS, it's a little bit different. As you can see here, there's no, like, chat and co work tab like there is on the ChatGPT iOS app. You have to press this little side panel right here, and you can see co work is right here.
So it is like and then you can create a new task in co work, and you can work within a project right here, and this is co work. It's kind of confusing. I'm not gonna lie.
Because, like, how do I even tell that I'm in co work right now? And then if I go to chats, it's just we have chats and then we have co work.
But when we wanna create a new task, like, there's no co work tab, it looks the exact same.
And, also, what's the difference between Claude, co work, and a normal Claude chat session? The lines are still pretty blurry, and it's very similar to the conundrum that OpenAI is in with the distinction between chat gbt and chat gbt work. And so it is a little bit confusing, but now you can access Claude Cowork, which is basically an agent running in the cloud.
You can ask it to do much more complex tasks that lasts for a really long time. And the coolest part is you can use the best model in the world, which is Fable five. And then one final little update for those of you who like using Claude in the terminal.
If you run terminal and you can now, if you run Claude, you can go slash config.
And then if you go search in the settings output style and then you press the right arrow key, you can now choose from different ways that Claude code speaks.
So you can have it be proactive, explanatory, learning, and this is a new feature that they added.
And yeah. So you can customize exactly how Claude talks to you.
And for the final update of the day, we have something that I think is a little bit slept on despite it getting 2,000,000 views. Directly inside Slack, they released Slack code launching today with agents from Anthropic, GitHub, Cognition, and Vercel.
This is an announcement directly from Slack.
If I need help, I can tag in a coding agent like Claude without having to leave Slack. I'll ask it to update our website with the latest designs.
But instead of having the agent in a regular channel, Slack does something new. It spins up a dedicated code channel. This is something we talked about this a few weeks ago with Buzz, which is a Slack clone
built for AI agents. I said the one thing I don't like about Slack is the agents that you use inside Slack, like with Claude Tag, can't spin up new channels. Well, now we're starting to get to that point where the new Slack code agents can spin up new code channels, which is really interesting.
I will say this isn't really for general, uh, agent use cases yet. This is more for developers, but I do think it's the beginning of Slack really allowing agents to go hard in these channels.
And I think that this is just the beginning of using AI agents a lot more inside Slack.
For this. And that channel has the right teammates and the right context from our previous conversations, and that allows the agent to get to work.
This isn't your traditional coding experience with one person and one agent working in a silo on their own. In the code channel, the whole team and the agent work together on the build. There's no specialized tools needed,
and it's all right here in Slack. And that's kind of the main value prop is that coding, especially like vibe coding or using AI agents to build an app, used to be this solo thing. Right?
You would work, uh, using your your the code would run locally on your computer. You would get an AI agent to make changes to the app. You would test it on your computer, and then you would make a PR.
Now you can actually prompt in a group setting with Claude, and you can kind of work with other people, and other people can actually see your prompts to the agent that will then build the app. And it is becoming a more collaborative process.
And I don't think that this is just gonna be for coding. I believe this is the future of all work with agents is kind of collaborating with AI agents and other humans to get work done at the highest level. And because talking to AI agents is a skill in and of itself, it's useful to do it with other people.
So other people can be like, hey. You shouldn't prompt it like that. You should try prompting it like this.
And I just think this is kind of the next iteration of agent native companies.
That means it develops, and the team steers the agent together. It's a whole new way to work in a true multiplayer experience.
Multiplayer agents, I think that's kind of the trend that we're in. I said this in the last episode.
First half of 2026 was about the personal agent, and I think the second half of this year is all about the transition from personal agent into creating a team of agents for an individual and then eventually a multiplayer experience where you have many humans interacting with many agents with a common goal of growing the business, which I think is really exciting and fun.
And there you have it. This was an agent native update. We talked about GPT messages.
We talked about GrokBot. We talked about Cloud updates. We compared the major super app platforms, and we talked about a new super app being implemented into Slack.
We covered a lot today. Thank you guys so much for watching. I do this every week.
Please subscribe. Please like. I'll see you here for the next one.
The Hook
The bait, then the rug-pull.
Riley Brown opens with a straight news hook: OpenAI just gave Codex the ability to read and send your iMessages, and that's only the first of five updates he's tracking this week across GrokBot, Claude, and Slack.
Frameworks
Named ideas worth stealing.
07:29model
The Super App Scorecard (GrokBot vs. ChatGPT vs. Claude)
Model quality: point to Claude/ChatGPT (GPT 5.6 Sol); Grok 4.6 judged weakest at writing and knowledge-work tasks
Browser: point to Codex/ChatGPT for its built-in in-app browser
UI ease of use: point to GrokBot for scoped, single-purpose agent 'bots' and simplest iOS setup
iOS app: split — GrokBot simplest to set up, ChatGPT most powerful (real-time voice can spin up a Codex session)
Knowledge-work docs (spreadsheets, presentations, front-end design): point to Claude
Riley scores GrokBot, ChatGPT, and Claude across five categories live on a whiteboard, landing on a split decision: no platform wins outright, so the right tool depends on the specific task.
Steal forevaluating any set of competing AI agent platforms on your own criteria instead of defaulting to one vendor for everything
CTA Breakdown
How they asked for the click.
VERBAL ASK
04:32product
“Grab HubSpot's free AI agent cheat sheet.”
Mid-roll sponsor read placed right after the Codex iMessage demo; Riley names the specific value (seven tools, setup time, cost, starter prompts) rather than a generic plug, with the real offer link in the description.
A week where xAI undercuts Anthropic and OpenAI on price, and GrokBot's named-agent structure hints at how every AI platform is about to fix the chat-versus-work problem.
An interview with an early power-user on why the open-source Slack-for-agents app is going viral, then a full walkthrough of running a personal AI agent team inside it.
Claude Desktop and the Codex app both quietly shipped native, multi-tab in-app browsers within days of each other — used right, one prompt now opens, drafts, and stages a dozen tabs at once.
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.
A 39-minute live build: one AI agent cloned from a template, one built from scratch, both running 24/7 in Slack and iMessage using the GRASP framework.
A 22-minute weekly roundup covering the open-source model that finally passed the vibe check, Codex screen-to-skill recording, the SpaceX/Cursor acquisition, and two Claude updates.
00:30
00:34
01:16
01:34
01:51
02:12
02:25
02:42
02:59
03:16
03:33
03:50
04:07
04:25
04:56
05:11
05:43
06:12
06:24
06:46
06:58
07:16
07:50
08:07
08:24
08:41
09:05
09:15
09:32
09:49
10:07
10:41
10:58
11:15
11:32
11:49
12:23
12:40
12:58
13:32
13:49
14:06
14:23
14:57
15:17
15:31
15:49
16:06
16:23
16:57
17:14
17:31
17:48
18:05
18:22
18:57
19:25
19:41
20:05
20:30
20:33
20:50
21:13
21:31
21:53
21:59
22:42
25:21
Video of the Day55:15
Video of the Day23:10
47:51
39:16
22:18