Big Idea

The argument in one line.

Claude Desktop and the Codex app both shipped native in-app multi-tab browsers this week, and treating that browser as the agent's action surface rather than just a display turns routine multi-tool work into a single prompt that opens, drafts, and stages everything across many tabs at once.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You already run daily work through Claude Code or the ChatGPT/Codex app and want to stop tabbing out to Chrome mid-session.
You do content, marketing, or research work that touches many web tools (Notion, Google Docs, X, email) in a single sitting.
You want concrete, repeatable workflows for skills, batch tab-opening, and running parallel agent research sessions.

SKIP IF…

You don't use Claude Code or the Codex/ChatGPT app at all — this is entirely about those two apps' new browser feature.
You're looking for a technical explainer of how the in-app browser is implemented — this is a usage walkthrough, not an architecture breakdown.

TL;DR

The full version, fast.

Claude Desktop and the Codex app both shipped native in-app multi-tab browsers within days of each other, letting an agent open and act across many web destinations without leaving the chat. Codex's implementation is further along: its browser opens before a chat even starts, its @-mention plugins actually trigger actions, and skills can be written to auto-open results in tabs. The video demonstrates four ways to put a tab in front of an agent — manual, single-open, bulk-open, and skill-triggered — then walks through real daily workflows: an email-draft skill, script-driven Google Images b-roll sourcing, batch tweet queuing with image uploads, and three parallel research threads run from one session. The framing throughout: stop telling agents what to do and start asking them what you should work on next, then let the browser surface everything needed to act.

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 →

Chapters

Where the time goes.

00:00 – 02:17

01 · Introduction

Cold open: Claude Desktop and the Codex app both shipped in-app multi-tab browsers on the same day. A live demo opens YouTube analytics, a Notion video database, X notifications, top emails, and a soccer score in one prompt using sub-agents.

02:17 – 03:43

02 · What's New in Claude Code

Claude Code's browser previously opened only localhost links; it now opens any live site. A new annotation tool, compared to Excalidraw, lets you mark up any open page and send the note into the chat.

03:43 – 07:40

03 · Browsing Inside Codex

Codex's browser can open before a chat even starts (Cmd+T), its @-mention plugins like Google Drive actually trigger actions, and it writes directly into an open Google Doc using full context on the user.

07:40 – 11:31

04 · 4 Ways to Use the Browser in Codex

A whiteboard breakdown of the four ways to get a tab in front of the agent: open it manually, ask it to open one thing, ask it to open many things, or bake the behavior into a skill.

11:31 – 14:41

05 · My AI Operating System

Riley reframes his workflow around a Jack Dorsey quote — ask the agent what to work on next instead of telling it what to do — demonstrated by prepping two upcoming podcast episodes and letting the agent open every reference tab needed.

14:41 – 21:34

06 · 6 Workflows I Use for the Browser Inside Codex

Walkthrough of daily workflows: an email-draft skill that opens each reply in its own tab, script-driven Google Images b-roll sourcing, batch tweet queuing with image uploads, and three parallel research threads spun up from one session.

21:34 – 23:10

07 · The Takeaway

Closing argument that a full in-app browser is required for an agent to do almost anything, a recap whiteboard of the whole video, and a call for viewer feedback on what to cover next.

Atomic Insights

Lines worth screenshotting.

Claude Desktop and the Codex app both shipped in-app multi-tab browsers in the same window, so an agent can open a dozen destinations without ever handing control to Chrome.
Before this update, Claude Code's browser could only open localhost links; it can now navigate to any live site.
Codex's browser can open before a single message is sent by pressing Cmd+T on a blank chat, while Claude Code's browser only activates after the session has started.
Codex's slash-command and @-mention plugins, like Google Drive, actually trigger actions, while the same @-mentions in Claude Code currently do nothing.
There are four ways to put a browser tab in front of an agent: open it manually, ask it to open one specific thing, ask it to open many things at once, or bake the behavior into a reusable skill.
An email-draft skill can read every open support ticket, write a reply for each, and open every draft in its own browser tab so nothing sends until a human clicks.
Feeding a script to an agent and asking it to open one Google Images tab per line turns b-roll sourcing into a single batch job instead of a manual search per cut.
A single Codex session can spin up three separate research threads in parallel — hook research, hardware shopping, and podcast-guest sourcing — each opening its own tabs and finishing independently.
Jack Dorsey's framing for agent-era work is to stop telling agents what to do and instead ask them what you should work on next, then let the agent open everything needed to act on the answer.
When an agent is connected to calendar, email, and notes, it can read all three and hand back a ranked list of what to work on instead of a person triaging each tool manually.
The video argues the browser itself is becoming the interface layer for AI agents, since it already sits on top of nearly every tool people use for work.

Takeaway

The Browser Is Becoming the Agent's Workspace, Not Just a Display

WHAT TO LEARN

When an AI agent can open, read, and act across many browser tabs at once, the browser stops being something you navigate and becomes the surface where multi-step work actually gets done.

02What's New in Claude Code

Claude Code's browser previously could only open localhost links; it now opens any live website, so an agent can research external pages mid-session.
A new annotation tool lets you select part of any open page, add a note, and send it straight into the chat, similar to a lightweight Excalidraw layer over the browser.

03Browsing Inside Codex

Codex's browser can be opened with Cmd+T before you've even sent a message, while Claude Code's browser only becomes available after the session starts.
Codex plugins can be triggered by @-mentioning a connected tool like Google Drive and it actually acts; the equivalent @-mention in Claude Code currently does nothing.
Because Codex already has memory, Notion, Drive, and task history connected, it can write directly into an open document with full context on the user, not just what's in that one file.

044 Ways to Use the Browser in Codex

There are four ways to get a tab in front of an agent: open it yourself manually, ask the agent to open one specific thing, ask it to open many tabs at once, or bake the behavior into a reusable skill.
The most durable of the four is the skill: once written, it fires the same open-tabs-and-draft behavior automatically every time the trigger condition is met.

05My AI Operating System

The organizing principle, borrowed from a Jack Dorsey post, is to stop telling agents what to do and instead ask them what you should be working on, then let them open everything needed to act on the answer.
With calendar, email, and notes connected, the agent can read all three and hand back a ranked list of what to work on next instead of the person triaging each tool manually.
Once a task is picked, the same session opens every reference doc, guest link, and related video needed to execute it, collapsing several manual tab-hunting steps into one prompt.

066 Workflows I Use for the Browser Inside Codex

An email-draft skill reads every open support ticket, writes a reply for each, and opens every draft in its own tab so a human only has to review and send.
For video b-roll, feeding the agent a script line-by-line and asking it to open one Google Images tab per line turns hours of manual searching into a single batch job.
The same batching pattern works for social posts: point the agent at a folder of images and it drafts captions and opens each ready-to-send post in its own tab.
A single session can spin up three separate parallel research threads at once, each opening its own tabs and finishing independently, for tasks like hook research, hardware shopping, and guest sourcing.

Glossary

Terms worth knowing.

In-app browser: A browser embedded directly inside an AI chat application so new tabs open within the same session instead of switching to a separate app like Chrome.
Skill: A saved, reusable instruction set an AI agent runs whenever a matching trigger phrase or task comes up, without the user re-explaining the steps each time.
Sub-agent: A secondary agent process spun up by a primary agent to handle one part of a task in parallel, such as opening one specific browser tab.

Resources

Things they pointed at.

00:00toolClaude Desktop / Claude Code

00:00toolCodex (ChatGPT app)

00:45toolNotion

04:10toolGoogle Docs

05:40toolX (Twitter)

11:51linkJack Dorsey tweet on agent workflows

21:37toolvMix

Quotables

Lines you could clip.

00:01

“OpenAI and Anthropic made the same update on the same day to the Claude desktop app and the Codex app, and I think this is gonna be massive for how we do business with AI agents.”

cold-open thesis, sets up the whole video→ TikTok hook↗ Tweet quote

06:41

“The browser is the most underrated interface for all AI agents. It already sits on top of nearly every tool we use, give it memory context and the ability to act, and it stops being a place you browse the Internet. It becomes a place where you get work done.”

standalone thesis statement, needs zero context→ newsletter pull-quote↗ Tweet quote

11:51

“I've shifted from telling agents what to do to asking them what to do and pulling the best thread.”

borrowed Jack Dorsey quote, reframes the entire workflow philosophy→ IG reel cold open↗ Tweet quote

22:17

“These browser updates are only two days old, so my workflows are going to be drastically changed over the next few weeks.”

urgency/recency hook, sets up a natural follow-up video→ TikTok hook↗ 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.

metaphoranalogy

00:00OpenAI and Anthropic made the same update on the same day to the Quad desktop app and the Codex app, and I think this is gonna be massive for how we do business with AI agents. So I'm talking about their in app browser. On both Quad desktop and the Codex ChatGPT app, you can open up multiple tabs in every chat session, which is especially useful if you use these tools for running a business, doing marketing, and content.

00:26Check this out. This is the latest update to Quad Code. And as you can see here, I have 10 different browser tabs open inside Claude Code.

00:36This is a brand new update. And this is the Codex app. So I can run this prompt right here.

00:42And I'm saying open my YouTube Studio Analytics, open my Notion video database, open the x my x notifications, my five most important emails, and the score of the Norway soccer game, and open all of these as new tabs in the in app browser.

00:59And I'm telling it to use sub agents. Check this out. And you can see the sub agents are spinning up right here.

01:05And look at this. All of these different browser tabs are opening up right here. We have my YouTube analytics.

01:11We have the Twitter notifications open. We have my video database.

01:15This is where I film all of my videos. And here, we have the Norway score, two to one.

01:21They lost, rest in peace, into Norway. And we can full screen this, and it should be opening up some more browser tabs. And look, all of these new tabs are opening up.

01:31So these are different emails. The most important emails that have been sent to me are opened up in their own tab. And then we also have a document that it created and opened in a new tab, which is called recent support ticket summary.

01:45And so as you could see here, these AI agents inside Codex and Claude Code can take actions for you and then open them up in different tabs. So in this video, I wanna first talk about the updates to the browser in Claude Code and Codecs. And I also wanna talk about why these tools are becoming operating systems and why browsers are actually so important for this new future of using these AI powered super apps.

02:11And then I wanna talk about some really cool workflows for using the browser inside Codex and Claude Code. So first, I wanna talk about the updates to Claude Code.

02:21So before three days ago, you could not open up this browser to anything except for a local host link. So if you're building an app locally, the app can open it in the browser, but you can't go to an actual site.

02:34Now, you can just ask Claude, I wanna write a script on the best ways to use Claude code. Can you please open up some links there really quick? Don't think for too long.

02:43Just just open up some links real quick. And you can see here it's opening up the best practices, and now we can scroll.

02:49Here it's opening up a new tab, and now we have two tabs open. What I don't like about this is it kinda does them one at a time instead of just doing them in the background. A few weeks ago, I made a video about why this is so important, and it's because I believe that it's kind of reinventing the way we use computers.

03:05Instead of navigating to ARC and then going to the tab that we need to go to, you can just go to an AI agent, you can start a new chat, and you can say, hey, I'm about to work on a video about this thing. Please open up anything relevant.

03:20And it can just open up all of that information for you. And, of course, I don't use Claude code for knowledge work. That's why I use Codecs.

03:29So here, it's actually not going to be able to do that very well for me. Another thing Anthropic added is this annotation tool, and this is one of the better annotation tools. It's kind of like Excalidraw, so it opens this up in the bottom.

03:42And no matter what page you're on, I could communicate to Claude. Watch this. I could select this part right here, and I could add text and I just say, do we really need this question mark?

03:54And then I can add it to the chat and you can see here that it inserted this into the chat right here. So I can communicate with Claude. And if I actually had Claude set up to my Google Docs, which I actually don't right now, I do have it set up on Codex, I could just ask it to remove it, and it would.

04:10And so now I wanna show you how to use this feature directly inside Codex. So if we go to a new tab, what we can do is we can press command t before we even type anything in, and we can just type in Google Docs. And we can just navigate to Google Docs, and we can just use it like a normal web browser.

04:26And remember, you can press this button to full screen it, and then you can actually lower this, and you can actually just use the browser full screen. And if you want to communicate with AI, we could create a new document right here.

04:37I could say Riley's Obsession with Super App Browsers.

04:43Browsers. And since Codex knows exactly what we're working on, I can just open this up right here and we can use 5.6 Sol. Hey, please use everything that you know about me, and then write two paragraphs on this.

04:55Don't think for very long, just write it out right now inside this doc. And I'm just gonna change it to something less smart because I do want it to go a little bit faster. And we can just fire this prompt off to Codecs and it will write directly on this document.

05:09And because this is the in app browser inside Codecs, it knows exactly what's open at all times, which is really cool.

05:17And it will just write to this. And because I use Codex for nearly everything, it knows me incredibly well. You can see here that the Google actually has AI built in, but this AI actually knows nothing about me other than what's in my Google Drive.

05:32Codex knows everything about me, basically. It has my Notion, my Drive, memory about all of the tasks that I do, literally everything.

05:40And there you go. Codecs just added this, and I can say first paragraph too long.

05:47Please format this like a tweet, actually.

05:52And remember, I can always press command t to create a new browser tab, and I could actually open up Twitter, and I could go to Twitter. But I wanna show you something real quick.

06:03I'm gonna close this tab, and there you go. Now I'm gonna say, okay, I want to tweet this from my agent native account, but it's not quite there. Can you please make three versions of this and then open all three as a tweet?

06:18Don't actually tweet it. Just get it ready for me to send, and I'll send the one manually that that I want to send. These should be open as new browsers inside the Codex browser.

06:27Three tabs. And so I can just fire off this prompt. AI is gonna make three different versions of this, and it's going to open them up in the browser, and I can just select which one I want to send.

06:41There you go. So it took forty four seconds. It said three tweet versions are open in separate Codex browser tabs.

06:47Prefilled under the agent native account, nothing was posted. So now I can just navigate to one browser and be like, this is actually interesting. I don't like it that much.

06:56I can just close that tab. We can go to the next one. The browser is the most underrated interface for all AI agents.

07:04It already sits on top of nearly every tool we use, give it memory context and the ability to act, and it stops being a place you browse the Internet. It becomes a place where you get work done. That's pretty solid.

07:15So I can go ahead and fire this one off and we can close these tabs. There's actually four main ways to use the browser inside Codecs and some of these will work with Claude Code.

07:28I've just noticed that it's much easier to use the browser inside Codecs. One example is that you can't like, if you're on a new chat, you actually just can't open the browser until the chat actually starts.

07:38Whereas on Codecs, if you create a new chat, you can just press command t and open up the browser. So they're actually pretty far ahead in terms of like making browser functionality actually useful, and their plugins are a lot better. Like I can at mention Google Drive to get it to control Google Drive and open it up in the browser.

07:56Whereas Claude, you type at drive, nothing happens.

08:00Or slash drive, nothing really happens. Their skills aren't synced from code to home, so that doesn't really make that much sense.

08:08And then if you go to customize and you go to connectors, you can see that I have set up Google Drive on Claude Code, I think.

08:16It's just a lot more confusing. So that's why I'm showing you inside Codecs, and I just know it a lot better, and it's much more useful for me for as a marketer at a startup.

08:25And so the way that I use the browser is you can just open it up manually, and you can do that with command t. This is really useful. Like, sometimes I'll literally just start a new chat on Codex, and I'll just go straight to the browser.

08:36And I can just navigate anywhere I would want to, and then I'll tell Codecs to edit it or do something or open up new tabs or take different action. And so the next way you can use Codecs to use the browser is you could just tell the AI to open up something in the browser. That's really easy.

08:52You can just go here, hey, we were working on a long form video on Devin. It's in Notion. Can you please pull that up in the browser?

08:59And so I can just fire that off. And since I have this Notion integration built in to Codecs, it can very easily find it and then open it in the browser. And there you go.

09:11It opened up the Devin Beginner's Guide and here's the intro. This is the video that I wanted open. The next way that you can use Codecs to use the browser is you can use AI.

09:21You can ask AI to open many tabs. I can go to a new chat session, and again, I can open up the browser by pressing command t. In this case, I'm just gonna say, hey, I don't know what long form video I need to work on.

09:33Can you open all my long form videos? And you know what? Open all my short form videos in the browser tab as well.

09:38I just wanna see all my videos that are in progress in Notion open up right here. And so I can just ask AI to open up a ton of different browsers. And there you go.

09:48As you can see here, literally all, like, 16 videos that are in progress right now are open right here. And this is actually this current video that I'm filming right now. And the last way to use the browser inside Codex is the most useful and the most fun, and that is putting it in a skill.

10:06So you can think of this as like any time x happens, open x in the browser. Let me show you what I mean by this. So in ChatGPT, now Codex, they've merged the apps together.

10:18I can go to a new thread. Hey, I need you to look at the support tickets right now, and I need you to draft a response to all of them. And I can go slash draft email draft.

10:27This is my email draft skill that I'm referencing. Part of this skill is to open these up in a new tab. So it's going to do that automatically.

10:38And as you can see here, it says I'm using the email draft workflow. I'll inspect the repository guidance, identify every current support ticket, read each full thread, save replies, verify them, and open each in its own Codex Browser tab. Nothing will be sent.

10:54And here we go. We see these tabs opening up. These should all be draft links.

11:00So not only can it search through my email and suggest responses, it can create a draft and then show me the draft right here. So this is a draft that I can just send or I can edit, and it did that for all of the important ones.

11:15Sorry to hear about this. Thank you for sending the receipt. I've flagged the $20 charge.

11:19Here we go. These are responses that are just drafted up for me to send.

11:25And this is an actual workflow that I use to basically send all of my emails. So those are the four ways that I use the browser inside Codex. Now I wanna spend a little bit of time talking about how this is becoming my personal operating system.

11:39And right after that, we'll get to six really cool workflows, but I wanna talk about this first. And so the best way that I can describe this actually comes from a tweet by Jack Dorsey.

11:48He's the old CEO of Twitter, and now he's at Block, the CEO of Block. And he said this, I've shifted from telling agents what to do to asking them what to do and pulling the best thread.

12:01And here's an example of this. Based on everything you know about me and the important deadlines coming up, think about email calendar, what what is the next task that I should work on? Or give me my three options real quick.

12:12And so I can just fire this off to AI. I don't need to, like, go check my email, check my Slack, check everything. I'm just sending AI to evaluate all of it.

12:21Right? It's connected to all of my tools right now. And then it's going to make a decision.

12:26Right now, it's reading the Google Calendar, and then it's gonna read Notion. It's gonna read my email, everything. And it's going to present to me what I should be working on.

12:35You know what? I'm gonna start on this one right here. Okay.

12:37Yes. The two podcasts that I need to record on Thursday are really important. Open up the Google Doc for both of those, and then open up any external, like, in the tab in the other tabs, any research on this podcast.

12:49And then in the chat, not in any of the docs, for each of these guests, I want you to suggest 10 questions that I should ask and any pointers on, like, how I should structure it for the most views. Right? So I just took it.

13:01I just told the AI to go look at all of my stuff, and now it told me that I should work on this. And so I'm like, yep. Alright.

13:08Let's do it. And now I'm basically pulling this thread, and now it's gonna open all of the tabs that I'm gonna need to do a good job at that. You know, in the previous world, I would need to, like, open my computer, process what I need to do, open all my tabs, open Notion.

13:24Here, we have an app that can just open all my stuff, which is a completely different way of interacting with a computer. And here we go. It's opening all the different tabs that I need.

13:33So here's the document for one of the podcasts that I'm doing, and we're gonna be giving AI agents the power of generating assets for this. And so it opened up the gen media. So this is something that I could read that would help me prepare.

13:46It also opened their LinkedIn of the guests that I'm bringing on, and here's the next doc of the next person that I'm bringing on. And it opened up all of the it opened up a podcast that he was on previously, which is incredibly useful.

13:58Now I can just listen to it before I go on the podcast, and that will help me create this document that I actually haven't started on yet.

14:07So I I need to prepare for this, and that's exactly what I can do. And so I can say, please, on the Steven episode, based on everything that you know about me and what I've talked about, my interests in terms of, like, using paper and stuff, please write out a good intro for this episode, and I'll edit it if I like it.

14:25And so now we can edit this document directly, and it can use this is using the Google Docs plugin, right, right, that we could at mention, or you can just speak it out, and it will do exactly that. Please actually put it in the Stephen Haynie episode doc.

14:39And there you go. It added the hook directly to the document, and we have the link here.

14:44If I didn't have this open, I could very easily just open this up in the browser, and then I could see the hook directly in the browser. And so that's basically how I use this as an operating system. I wanna finish this video showing off some workflows for using the browser inside Codex that I think you'll think are really cool.

15:03And so the first workflow, I actually already showed you. And this is just my email draft skill where I will tell Codex to draft an email to all of the important emails that I have in my inbox, and then it will open the draft to the response in a new browser tab where I can simply respond. Okay.

15:21So the next one involves Google Images. And now this one is a little bit more niche to me because I use it for content creation, but let me show you this workflow. So I'm inside Codex, and I'm working on this script.

15:33And I have this video style where I record a video and I have images pop up on the screen. And for this, I actually have to manually go search Twitter, go search Google Images to find the images that I want to show on screen. Well, now I can do something like this.

15:47Hey, I wanna find at least 15 images to show on screen during each line. What I want you to do is open up a open up browser tabs in order of where they would appear in the video and search relevant things that would show up on the screen on Google Images. So I want Google Images open so I can just go tab to tab and see what you searched and then look through those images so I can very easily add them as b roll in this script.

16:11And so now Codex is going to analyze this script every single line, And then it's going to decide what it should search for on Google Images, and then it's gonna open all of them up as a new tab. And there you go. Look at how fast it does it.

16:25Bang. It's already done. And I love that it can just do it in the background.

16:28OpenAI GPT 5.6 sole model announcements. Now I can just go straight to the model announcement. Here we have this image that we could choose from.

16:37Here we have the OpenAI logo, new AI model launch, and we just have all of these. Here's the Claude Fable.

16:45I literally included this. If you watch my content on short form, I would include this image. Here we have Anthropic CEO, AI safety warning government.

16:54These are all relevant images that I would use as b roll in my video. And so the next one, I actually already showed you, but I wanna show you another example of doing tweets.

17:06And this involves uploading it with assets. So what I can do here is I'm actually gonna go ahead and, hey, I want you to queue up tweets.

17:16So each tweet should be a browser tab open, ready to tweet.

17:21I should be ready to press send. And I want you to look in my downloads. I have images that I wanna tweet, and I want you to analyze the images and come up with the caption for these images.

17:31Each one should be on its own browser tab. And, yeah, just get them ready to to send, and I will take a look at them each in their own tab.

17:39And so Codecs can just go into any file on your computer, and it can upload them to Twitter or at least get them ready to be tweeted each on their own browser tab. And it can even analyze the photo and come up with the best caption for it. You could do the same with Instagram as well.

17:55And if you wanna wait for it, remember you can press command t to open up this side browser. And there you go. It wrote the caption.

18:03It should insert the image. Let's see. And you can see here it's actually manually going through and uploading it.

18:10It's choosing the image file from my downloads. And there you go. Here's my latest screenshot from my best performing YouTube video.

18:18And now it's gonna do it again in the next tab. And the next tab, it's doing the same thing. And these are very similar thumbnails.

18:25They're not exactly the same, but it's taking my last 10 images and getting them ready to send. And obviously, this isn't how I would actually do it. I would probably have a separate folder for each post, but you can just tell Codex to go find any folder on your computer and queue them up to be posted on any platform.

18:41So we have this one and this one. And this is pretty cool. Finally, we're gonna do three prompts at the same time.

18:46We're gonna do one for hook analysis. We're also gonna do another one for product research and podcast guests. This is gonna be really cool.

18:55We're gonna do this from one codec session. So what I'm going to do here, I'm going to say, hey, I want you to create three separate codecs threads, all of which should open up many browser tabs.

19:07So this should create three sessions. In each of these sessions, you should open up many browser tabs. For the first one, I'm gonna say, look up the hooks of the 10 videos for my next video and please can you make a Google Doc on the first tab and then open the videos in each of their own tabs inside Codecs.

19:25The next one will be more relatable. For number two, I'm going to say, I'm looking for an upgraded Windows PC that is really good for filming with multiple cameras and using vMix.

19:37I want you to do in-depth product research and open all the relevant links in their own tabs. And for the third one, I want you to do research, in-depth research, and find 10 ideal podcast guests.

19:48And I want you to open up links in their own browser tabs that shows me exactly why you think they would be a good guest. And if it's a YouTube video, a link to the direct timestamp, and that's exactly what I want you to do right now.

20:01And so watch this. I'm going to fire this off. It's going to automatically these should be their own sessions, not inside a folder.

20:10And so when I say that, it's just gonna put them in the uncategorized thing. So it's gonna create three new chat sessions right here, and they're gonna show up right here, each of which will spin up browser tabs.

20:22And here, we created one, two, and three research sessions. So I can click on this, and we can see here that here it's looking for hooks relevant to YouTube videos for Riley's next video.

20:34On this one, it's researching conduct in-depth research for an upgraded Windows PC that's excellent for professional filming with a bunch of more details.

20:44And then it's also researching podcast podcast guests. And take a look at this. All of them are done, and we can see that it researched vMix PC options.

20:53It not only created a doc for me to look at with all of the different links and examples here. I also see a ton of tabs open. If I wanna click on any of them, I can do additional research.

21:06So this is just a really good way to do product research for anything that you're trying to buy. Here, it did research on podcast guests. And here is the completed Google Doc for the hooks that it created.

21:17I can open this up. It did hook research, and it opened up all of the YouTube videos that I should check out.

21:24I can see their hook directly in the YouTube video. And then for the guests, we also we have all of these different people that I could reach out to, including Nate b Jones, which would be a great person to have on the episode, as well as Nate Herc. And then it opened up all of the relevant tabs up here, and it linked to one of their videos.

21:41Anyway, long story short, I believe that in order to have an app where you can use AI to do almost anything, you need to have a full browser. I think it's really annoying when you're using an agent tool and it spins up an external browser and takes me to Chrome, takes me off the platform that I'm working in. I just want it to open up in the app that I'm working in.

22:02And both Codex and Claude desktop are adding this feature directly inside.

22:08And so we talked about the different ways to use it. We also discussed why I believe it's becoming like an operating system, and I showed you many different workflows. And honestly, these browser updates are only two days old.

22:19So my workflows are going to be drastically changed over the next few weeks because of these updates. And so I'm curious, please let me know in the comments what workflows you're using that involve the browser. And then anything else you think is cool right now, let me know what I should make a video on next.

22:36There's basically so many things to cover just inside Claude Code and Codex, and I want your feedback because we have a lot of fun episodes planned. And I thank you guys so much for watching.

22:46We finally have our new studio set up, and I feel like we're hitting our stride. I would love to get to four or five videos per week because there's just so much to cover right now. And all I really care about is helping people become agent native and use AI agents in their business to complete their work, to do better work, to do more creative work.

23:05Thank you guys so much for watching. My name is Riley Brown. Peace.

The Hook

The bait, then the rug-pull.

Two AI labs shipped the same feature on the same day: a full in-app browser. The video argues that's not a cosmetic update — it's the browser becoming the surface where agents actually get work done.

Frameworks

Named ideas worth stealing.

07:40list

4 Ways to Use the Browser Inside Codex

Open it manually with Cmd+T
Ask the agent to open one specific thing
Ask the agent to open many tabs at once
Bake it into a skill that always opens tabs on trigger

The four ways to get a browser tab in front of an AI agent, ordered from most manual to most automated.

Steal forany AI agent workflow that has to touch multiple web destinations

14:41list

Workflows for Using the Browser Inside Codex

Email Draft Skill — auto-open drafted support replies in their own tabs
Google Images B-roll Sourcing — open one tab per script line with relevant image searches
Tweet Queuing with Image Uploads — draft and open multiple ready-to-send tweet tabs from a downloads folder
Parallel Multi-Session Research — spin up 3 separate Codex threads that each open many research tabs at once

The concrete daily workflows Riley demonstrates on top of the in-app browser.

Steal forcontent/marketing ops that involve repetitive research or drafting across many tabs

Storyboard