Transcription
Anthropic just released Opus 5, and in many ways, it's better than Fable, and it's only half the cost. We're going to talk about this today, first.
But what's even cooler than the model release, in my opinion, Anthropic just launched a voice mode in their iOS app that can be used as a real-time voice agent with Opus 5. And you can connect this voice agent with all of your existing apps and tools.
But 10 minutes before this voice launch, OpenAI released their voice feature inside Codex that allows you to control Codex with your voice in real time. And so I want to talk about these updates today, too. We're also going to be covering all of the previous product releases by Anthropic and OpenAI over the past few weeks. And I'm going to tell you how I plan to use all of these features inside my startup because that's what we discuss here on Agent Native.
Okay. So, 5 hours ago, as of when I'm recording this video, Claude introduced Opus 5, and they described it as a thoughtful and proactive model that comes close to Frontier Intelligence of Fable 5 at half the price. So, what I find to be interesting is that Anthropic used this language right here. They said that their model Opus 5 is close to the frontier intelligence of Fable 5. Yet, they posted the model comparison card, and as you can see, it clearly is better in all but four categories. They said it's best in agentic coding. They also said it's best at knowledge work, which in my opinion is the most important part. They said it's best in novel problem-solving agentic search. It's not quite better at multidisciplinary reasoning. It's better at computer use, which is very useful, especially if it's half the cost because when you get into computer use with Fable, it gets extremely expensive. It's better at agentic coding, business workflows, not legal or health, but it's also better at biology. I can't really tell the difference between this model and Fable. It just feels kind of worse.
But someone did actually get early access, and that is Dan Shipper, and he has an opinion that I trust a lot when it comes to AI models. The first thing that he said is that this model is very hard to love. He said that, "We spent the last week at every testing it across coding, writing, knowledge work, and our internal agent. It argues with instructions, stopped before work was finished, and generally didn't play well with our existing skills and plugins like compound engineering. Our first reaction was, 'What have they done to my boy?' Then we deleted our existing skills and started from scratch."
This right here is why I think many business owners and managers and operators and people who do knowledge work are wasting so much time testing out the new models and the newest and latest models. If every time a new model comes out, you have to redo all of your skills, it is a complete waste of time. And he ultimately went on to say that it is a poor man's Fable. It has many of Fable's personality quirks without its like true genius.
So now when you go to Claude desktop app, whether you use chat, whether you use Co-work, or if you use Claude Code, you now have access to a new model, which is Opus 5, and you can use Opus model 5 and test it for yourself. So Anthropic is actually positioning this Opus 5 model as its everyday model for enterprises, knowledge workers, and developers. And in their marketing material, they mention knowledge work much more than coding tasks. So for that, I think the best way to test it would probably be in the home tab using Claude Co-work.
So to test it, I might do something like this. Make sure that you switch to Opus 5. I'm going to say, "Please extract the chorus.com site. Um, this is to get the brand. You can use Firecrawl to extract the brand of any website. Make a professional slide deck for investors to raise an A-round. Make up fake numbers. It should be a final deck that will get the job done. Make it amazing."
So, now we're going to be using Co-work and we're going to fire off this prompt. And there we go. Yeah, I'm not going to lie. I regret doing this prompt. I entered this prompt 41 minutes ago, and it's still going. This is insane. Oh my god, finally. Okay, so after 40 plus minutes, it's done with the presentation. And okay, I will say this is actually insane. Um, this is genuinely one of the best PowerPoints I've ever seen. If you go to chorus.com, look at the landing page. Like, look at the branding here. And then look at the presentation. Like, this is very similar. And these slides are like pretty damn good. Actually, it even pulled all of these different like logos, which I really like. It just took forever. And so, this model's really good. I recommend trying it within Claude Co-work or within Claude Code, but I'm not going to spend too much time on it because I want to get to what I think is a much bigger update, and that is the new voice mode inside the Claude iOS app.
So, this was released yesterday. So, as of yesterday, you can press this button right here, this white voice button. And as of right now, you can actually use it with Fable. I did not think you could do this. I actually don't think you could yesterday. You can see the model right here. And you can actually select. For the sake of this video, we can talk to to Opus.
"What's going on, my man?"
"Not much. Just materializing into your video, apparently. Sounds like you're demoing something live. What are you building over there?"
"Yeah, I'm actually building something. Um, there's a Notion document um that is called, what is it called? Agent Native Weekly. Um, and I created it on the 22nd. I'm filming it right now. And it talks about Opus 5 um and the Anthropic uh voice mode. What else are we talking about in this video? Can you search Notion and tell me?"
"Got it. Found it. So beyond Opus 5 and the voice modes, you've got a section on Codex and Claude copying each other, whether Chinese models get banned, the Kim K3 launch, whether software is dead, and whether we're already in the singularity."
"Yep, thank you for that. Actually, you know what? Are you able to edit this Notion document?"
"Yeah, I can. I've got the update tool for Notion pages, so I can add sections, tweak wording, whatever you need. What did you want changed?"
"Yeah. So, actually, um, you notice in the Anthropic OpenAI voice mode section, there's a tab in there called Claude Voice. It's empty right now. Can you please um do some research on the new Claude voice and write out like seven or eight concise bullets on what it can do and use cases and have hyperlinks to any sources you may use there. Where it says Claude Voice, I just removed the bullets there. I want you to just add them in the Claude voice tab."
"There you go. Do you see that? It just added this information here. Nice job, Claude."
"Ah, nice. Glad it."
"And you know what? At the bottom of this, you know, so we can go all the way down and we see what does this mean for business. Below this, I want you to add a new section in tabs of whatever you think I might want to include in the video."
"Done. I added a section called Your Move This Week with four tabs. Do this Monday, What to Watch, My Take, and Outro."
"Wow, that's really cool. Moving on. Can you please like um tell me a little bit about my emails? Like give me the latest update. Don't give any names of companies or people. Just tell me an update on what's going on in my email."
"Sure. Keeping it anonymous. The stuff that actually needs you. Two failed or overdue payments on subscription tools, so a card on file has probably expired. Beyond that, a warm follow-up from someone you met with today who wants coffee in New York next week."
"Can you actually draft up an email to emily@notumber.com? Just create a draft for me so that when I'm at my computer, I can send it letting her know that this video is complete that we're recording and it's the one that matches the Notion doc."
"Done. It's sitting in your drafts. Subject line is Agent Native Weekly Recording Done. And I included the link to the Notion outline so she can match them up."
"Hey, I actually want you to create a little landing page for me, Riley Brown, and this should just be my personal landing page. Can you please create it and then deploy it to Vercel for me?"
"Ooh, fun pivot. Love it. Quick question before I build any."
"Yeah. Um, in terms of Yeah, I want it to be like I have the Agent Native podcast, the number one place to learn about AI agents. Please make it very cool."
"Love it. Let me pull up the deploy tool and build you something slick. Let me just build it. Give me a moment to put together a cool."
"Okay, good. No need to say anything. Okay, as you can see, it kind of pauses and freezes."
"And deploy."
"See, for whatever reason, it does this like pausing thing, which is kind of annoying, but I think they're going to patch that pretty soon."
"A badge that says,"
"Okay, so you It's live."
"Um, that's pretty cool. Where's the link? The link is a bit of a mouthful to say aloud. So,"
"Okay, so this is a good learning moment. What you need to do whenever there's a link that you want, it actually can't do it in voice. You need to get it to type it out. So, you can exit voice mode and then you can say, 'Link here, please.' It just can't do it while on voice mode. So, if you ever need a link from the voice agent, you have to exit voice mode and say, 'Where's the link?' Okay, here's the public sharable link. Let's open this up. And there we go. It has this clearly vibecoded podcast, but the point is is you can vibe code in real-time voice on Claude."
But Anthropic wasn't the only one to release a cool new voice feature. OpenAI also released a voice feature directly into the Codex app or the ChatGPT Codex app that they combined, and you can now create a new voice chat. So I'm on the Codex desktop app here.
"Hey buddy, I need you to look at my drafts in my email. I I created one a little bit ago, uh, sent to Emily. I need you to pull that up in the Codex browser, please."
"Sure. I'll find that Emily draft and pull it up for you."
"All right. So, thank you. Um, you pulled it up for me. Thanks. I'm going to go ahead and send that."
Another thing that it can do is it can design directly on a Canva like board. For this, I use an a tool called Paper. But check this out.
"Hey buddy, old pal. I'm curious. Do you see my open Paper board right now? I'm I'm using Paper. You should have access to a Paper MCP or tool. Do you see the board? It says something on there. Please tell me what it says."
"Checking that now. It says 'Hello Monkey'."
"I know, but you you cheated. You're supposed to use the MCP. Can you use it? Like, can you add please add a uh little landing page? Actually, you should have in a previous chat, you should have access to an Agent Native site that we created. I think it was yesterday. Can you please add that to the board here? I want to redesign it."
"Okay, I'll pull that together. I'm using the Paper integration and finding that Agent Native site. Now I'm drafting a compact editorial landing page reference with a fall gray background, deep navy text, and the single electric blue accent."
"Wow. And you can see here the agent is editing it. This is pretty cool. I actually want you to create two more variations of this. Make them in different styles, please."
"Sure. I'll add two more. Working them in Paper again. Actually. And because you're you're not that strong of a model, can you actually delegate this to a more powerful model? Like spin up two more chats to do this. Keep the ones that you're doing now. Don't remove these. I want you to spin up two more chats to create them with a better model."
"Let me see what I can run in parallel. The magazine and terminal directions are on the board, and I'm keeping the original. I'm also spinning up two high reasoning tasks for more variations."
"Thank you. I've added a fourth direction as a high-resing task and I'm keeping the three existing boards intact. The two new high reasoning tasks are live and I don't see them remain untouched. I don't see them on Paper though."
"Checking that now. Right. You're right. Those two tasks are still running. So, their variations are not on the board yet. You should see them appear as separate art boards once they're finished."
"Art boards. It's It's They're not going to be on the same board."
"Checking that. Now, they're all going into the same gentle moon Paper board."
"Art boards inside it."
"There it is. Very yellow. Okay. The yellow um I don't like the one that says 'Your work has new superpowers.' Like that whole direction is very boring to me. Can you please just like adjust it to just make it like have more energy? Like make some graphic on it. That's cool."
"Got it. I'll adjust that. I'm reworking the yellow terminal one into something bolder and more kinetic, not just a copy tweak."
"Ooh, very nice. Very nice. That's a good direction. And yeah, look at this. We are just freely creating here with our voice. And this is just so fun. This is like talking to Jarvis."
"Done. The dark version now sits right beside the yellow one with a night navigation feel and a glowing route graphic."
"Very cool. Thank you."
Okay, so the video I did before this, I went much more in detail using this feature. I don't want to spend too much time on it, but I do want to show the other thing that OpenAI did is they actually allow you to do this from your phone. So I can click on remote here. So this remote is the way that I can use Codex from my phone. Meaning I can control Codex that's running on my desktop from my phone. For example, before I could select which computer, and this is the computer that I'm using, this Mac right here, and I can change this to be a normal chat and I can say, "Hi." So when I type in this prompt, it'll show up right here. You'll see that right now. Where is it? Oh, it's literally here. Yeah. So I can chat with it and I can say, "Are you connected?" And the point is is that these are like in perfect sync.
But now there's a new way to use Codex from your phone, and that is with real-time voice. So I can spin this up, and this isn't like a chat thread. All this agent can do is spin up new tasks. I'll show you exactly what I mean. I do want to like go through and clear these real quick. Let me show you how this works. So, I'm going to go ahead and use real-time voice. It's connecting to my Mac.
"Hey buddy, are you there?"
"Yeah, I'm right here."
"Hey, I I need you to just like go through all my chat sessions from the day and I I need you to spin up a new task and make a summary of all the things that I talked about today and then give me an include calendar in this. I just kind of want a summary of my day. So, go through all my Codex chats and such."
"Sure, I'll take a look."
"Very cool. I can see it's working in Codex. Can you spin up a new chat which gives me a little summary of my emails for the day? Don't include any product stuff, just like sponsorship stuff."
"Okay. Narrowing it to sponsorship only. I've created a separate one-time task for sponsorship brand deal and paid media emails only. It excludes product mail."
"Wow. Thank you. Do you see what tools I have or like skills you have access to on Mac right now?"
"Checking that now. Yes, I can see your available Mac skills and connected tools and your July 24th recap's ready too. You made progress across Notion research, audio setup, Agent Native design, and automations."
"Very cool. You know, you're good. You're good. One thing I want to discuss. Yeah, I just tell me the skills that you can see on my Mac."
"Checking what's available on your Mac right now. I can help with Paper, Gmail, Calendar, Drive, Notion, Slack, browser control, image and video work, social posting, PDF, stocks, and sheets."
So, as you can see here, I can fully control my Mac, right? So, when I go home for the day, I leave my computer at work, but I can always access it through this remote control. And before I would use this remote control, but now I can just speak to it. It is pretty insane that I can do real-time voice and talk to my computer even if I'm across the country.
So, those are some really cool updates to both Anthropic's tools and OpenAI's tools. And I want to dive a little bit deeper into this because Claude and Codex are seemingly at war, right? The the two front runners for super apps, whether it's Claude's desktop app or Codex or ChatGPT's desktop app, they're adding features basically every few days, and it feels like they're just copying each other. And I actually want to explain why this is a good thing for us. And by us, I mean a guy who wants to use AI agents effectively in my business, right? That is the ultimate goal.
And um the first or the first of four ways that these two are copying each other are the browser updates. So I've talked about this a lot. I've made a video on this specifically. Um if you go to the Claude app, you can actually just spin up a browser. And this is only in the Claude Code tab, right? If we go into any chat here, we can open up the browser by clicking this button. And so now this is a full browser, and you can open up multiple browser tabs. Same thing on Codex, right? I've talked about this a lot. You can open up a browser, and we can go to any website, and we can open up many tabs, and we can ask our agent to open up browser tabs. And so that is what they're copying, right? Claude added this feature very quickly after OpenAI added this feature. And that's just one of the features because these two have been copying each other in terms of the three tabs within the app. And so basically these frontier labs are trying to figure out how to stuff all of the things that you can do into one app, which is why I call it a super app. And it comes down to three things, right? You have chat, Claude Co-work, and Code. And on Codex, you have Codex, and then you can switch to ChatGPT mode, which has chat and work. So they're the same three things, right? You at the end of the day, there's chat tasks, which are like quick one-off conversations, work tasks, which are create any type of document, do any type of research, reach out to people, control email, use Notion, etc., and then you have code tasks. And all of these companies, they're seemingly going back and forth trying to figure out the best placement for these three tabs, and they just seem to be copying each other, and it's changing like every single week.
Also, the iOS simulator. So In the Claude app, you can actually open up the iOS simulator directly in the app. Now, if you ask it to create a Swift app, just ask it to open up the iOS simulator. In fact, this "Hello World" iOS app used a simulator, and you can actually see it right here, and it will actually open up the simulator. And when you tell it to run it, it will run it, and it will actually show up here in the simulator. In the Codex app, you can do the same thing by using this "Build iOS App" skill. This will automatically create the same workflow where the iOS app will open in your browser. So, this is an easy way to create iOS apps.
And then the fourth and final way that these two are copying each other is the record and replay. I've mentioned this in multiple videos before, but in the ChatGPT app or Codex app, you can use this skill called "Record and Replay," which allows you to just tell it to record your screen and then turn it into a skill. Well, Claude just added the same thing. If you go to a new chat, and I believe you have to do it on the home tab, and you can do it in like Co-work sign, and you can say, "Record a skill," and you can start recording. And so now it's recording my screen. And notice here it has capturing, and it shows zero steps. So if I were to want it to make it like uploading a YouTube video, I would say, "Go to YouTube." I don't know where it's getting all these steps from, which is kind of weird, but I could go to "Create Upload Video," and I could show that I can upload a video like this. And the point is is I can record my screen, and this context will immediately be uploaded into Claude. And the workflow I think is slightly better on the Claude desktop app. I can like record this here, and Claude will analyze it. And so I can say, "Please turn this into a skill called YT Video Upload." And this will allow me to take whatever steps that I do on my computer and turn it into a skill because remember, Claude and Codex can fully control your computer. And it's incredibly powerful. And it's really nice to be able to turn that into a skill. And it's way better than writing out the steps. Just show the agent exactly how to do it.
And so what does this all mean? And I think it's important to zoom out a little bit. Obviously, I like to talk about the day-to-day updates of these super apps, but it's important to like take note of like what the ultimate goal is of these labs. And I think this Greg Brockman tweet really sums it up. It's like, "One of the best features of ChatGPT Work is that it runs in the cloud, meaning that it works from mobile and when your laptop is closed." So what these companies are trying to do is they're trying to take all of the power of Claude Code or OpenClaude or these really powerful agents like Codex, and you're they're trying to give it to everyone. They're trying to get millions of people to use AI agents effectively. And the way that they're doing this is they're creating these co-work environments so that you can use these tools directly from ChatGPT. And so the one thing that I know for sure is it's going to get incredibly easy to use all of these AI agent tools. And all you'll need to know is how to set up skills or how to set up workflows, and they're going to be accessible from any device. Whether you're using the desktop app or if you're using the web app, you can select work, and you can basically do any tasks that you would need to do on Codex. That is the ultimate goal. You know, they're trying to take all of the things that everyone loved about OpenClaude and make it as easy to use as ChatGPT. That is the ultimate goal because on the phone, you can do the same thing. There's this toggle. You can toggle it to work. This is literally Codex. Like it's Codex light. Every time you use it, it spins up a little computer. And so if you need it to do a coding task, it will do a coding task. It can code anything. It can upload it to GitHub. It can deploy it to Vercel. You can have it control your emails. You can have it create a document. Anything you might want to do, you can do inside the work tab. And not a lot of people realize this. And you can do this on Claude as well. This is not just about ChatGPT. My point is is these companies that have millions of users using their app, they want you to be able to use agents because agents use up a lot of tokens, and they want that revenue. But that's not necessarily a bad thing because using agents can be incredibly useful for business.
So my main takeaway is as follows. Both companies are building something similar, and you shouldn't worry about picking the right tool or the right model. Like Opus came out today, and I do want to talk about it because I think it's good to know that it came out. But don't don't spend too much time on selecting a tool and selecting a model. If you use Claude or OpenAI's uh ChatGPT app, just go super deep and focus on your workflows. You're not going to pick the perfect tool. It just does not exist. And not to mention Cursor, they're going to be releasing their knowledge work version of their tool really quickly. Uh Gemini, they're going to make a competitor, right? It's hard to tell if it's AI Studio or Anti-Gravity, but you know that these are not just going to be the only players. Not to mention, there's also some people in China doing some cool things, right? Zcode, the people who created GLM 5.2, an open-source model. They have Zcode, which is their own super app, and it's very similar to Codex. It's actually frighteningly close to Codex. Uh Kimmy K3 also has Kimmy Code, which is a very similar idea. And so selecting the right platform is far less important than just focusing on your workflows as a business trying to use these tools effectively.
And if there's one thing I want to leave with you today is just figure out how to make your life better with AI. And that's probably means taking the tasks that you do every day and doing them faster or doing more and doing them better. Use AI to do more and use AI to do better. And if you're not doing that, then there's no real point to using AI other than entertainment. And it's a very expensive hobby. These tokens are not cheap. And so that's what I'll leave you with is like although we talk about these models and we talk about all these updates, I like to put everything in context to make you realize that at the end of the day, you need to learn the skills. And that's why I make all the other videos that I do. I try to create uh ideas and show you examples of ways that I actually get value from these tools because that's what you need to be doing. You need to figure out how to get value from these tools and only focus on the value, not the tools themselves. That's where all the value will come. And so that's what I want to leave you with today. Uh my name is Riley Brown. Thank you so much for watching. If you could like and subscribe, that would mean the world to me. I would love you forever. It helps me keep doing these videos. And anyway, this was an update. I'll see you here for the next.