Episode Categories:
Resources:
Join the discussion on LinkedIn: Got something to say? Let us know on LinkedIn and network with other AI leaders
Upcoming Episodes: Check out the upcoming Everyday AI Livestream lineup
Connect with Jordan Wilson: LinkedIn Profile
Start Here Series in our Inner Circle Community: Join for free access
AI Business Insights: Anthropic and OpenAI Updates, Meta’s AI Zuck, Salesforce’s Agentic Shift, and Strategic Takeaways
Recent weeks in the AI landscape have seen a flurry of targeted product launches, operational strategy shifts, and executive decisions. Specific upgrades and internal maneuvers by market leaders, and their direct impacts, provide instructive signals for executives seeking leverage in their own technology strategies. This overview extracts practical, business-relevant value from the latest movements at Anthropic, OpenAI, Meta, Salesforce, and others.
Anthropic Claude Updates: Automation Workflows and Business Productivity
Anthropic condensed the equivalent of multiple quarters’ worth of development into a single week. The launch of updated automated “routines” in Claude Code’s desktop app is particularly relevant for business users focused on efficiency. These routines allow users to orchestrate automated, repeatable tasks – either on local machines for maximum control or remotely via cloud execution that connects to SaaS platforms such as Gmail, Slack, and Google Drive.
With drag-and-drop interfaces and integration of parallel asynchronous jobs, technical barriers are reduced. The routines enable business leaders to systematically trigger deployments and automate multi-step processes, saving time on operational tasks. Users can also access side-chat for targeted code queries without disrupting workflow, which streamlines technical support cycles for teams where technical and non-technical staff collaborate.
Key Business Value:
Lower manual workload via automation of recurring workflows
Streamlined collaboration between technical and operational teams
Enhanced integration with existing SaaS investments
Claude Opus 4.7 Model: Benchmark Performance and Vision Capabilities
Anthropic’s release of the Opus 4.7 model introduces a measurable step up on SuiteBench – a key benchmark for coding effectiveness. Scoring 87.6, Opus 4.7 demonstrates enhanced capabilities in both software engineering and document analysis, with tripled vision performance compared to prior models.
This translates to more robust performance for tasks like document intake, spreadsheet automation, and visual data processing, specifically in business environments that rely on AI for document-driven workflows or client reporting. The “adaptive thinking” mode ostensibly allows models to dynamically adjust reasoning depth, offering the potential for more scalable, context-adjusted problem solving.
Key Business Value:
Higher accuracy in technical AI tasks, such as code review and report parsing
Greater flexibility in bringing vision-based analytics into product and operations
Early access to models that can self-regulate, potentially improving cost-efficiency on compute tasks
Claude Design: Rapid Prototyping and Exportable Asset Creation
Anthropic’s new Claude Design tool, in research preview to paid users, offers the creation of editable design assets—including UI prototypes, slide decks, and even animated videos—from simple text prompts. The assets can be exported directly into multiple formats (HTML, CSS, React, PDFs, PowerPoints, and Canva), and designs can ingest brand libraries such as Figma to match an organization’s visual language.
For organizations seeking rapid prototyping or internal enablement, this bridges the gap between concept and presentation-ready deliverable, freeing teams from complex software or specialized design staffing for early-stage assets.
Key Business Value:
Accelerated creation of client-ready or board-ready visuals from standard prompts
Lower overhead for prototyping and design iteration
Direct alignment with brand standards without extra manual adjustment
OpenAI Internal Revenue Dispute and Strategic Positioning
An internal OpenAI memo surfaced that disputes Anthropic’s revenue reporting, with OpenAI citing an $8 billion discrepancy based on gross versus net revenue calculation. The underlying business takeaway is the importance of transparent and standardized financial metrics—especially for vendors and partners evaluating market health and potential for long-term collaboration.
The memo also notes OpenAI’s foundational partnership with Microsoft has created limitations on enterprise reach as more businesses turn to AWS platforms. This reveals the current strategic importance of diversified infrastructure partnerships over reliance on a single cloud vendor—an insight relevant for scaling SaaS solutions.
Key Business Value:
Identifying reliable financial health metrics among AI vendors
Understanding the implications of infrastructure dependency when integrating major AI platforms
Benchmarking competitive posture ahead of potential public listings and enterprise procurement cycles
OpenAI Leadership Debate and Strategic Focus
Amid OpenAI’s approach to IPO, concerns from shareholders regarding executive decisions—specifically related to side investments and focus—highlight real risks tied to leadership distraction. The company’s move to direct employee focus on core products (and away from side projects) illustrates an operational discipline required when occupying a market leadership position.
Key Business Value:
Assessing executive alignment and risks in technology partnerships
Recognizing the necessity of focus and product clarity in rapid-growth AI environments
Mitigating regulatory and conflict-of-interest considerations for technology vendors serving enterprise needs
Salesforce Headless 360: Agentic Shift in Software Usage
Salesforce announced Headless 360, introducing more than 100 new AI-powered tools for agentic automation within their ecosystem. The initiative boldly questions whether CRM software in the future will serve humans or AI agents acting on behalf of the business.
With agents already performing 30% to 50% of Salesforce tasks, the emphasis shifts to product strategies that anticipate autonomous usage—diverting human capital toward higher value automation and prompting organizations to design workflows compatible with machine-to-machine integrations.
Key Business Value:
Strategic planning for SaaS deployments where agents, not end-users, are primary actors
Evaluating software purchases in terms of agentic compatibility and interoperability
Prioritizing staff upskilling in workflow design and automation oversight, rather than routine usage
Meta and Executive Digital Clones: Leadership and Internal Engagement
Meta's development of an AI-powered digital “clone” of its CEO—trained on communications, strategies, and organizational data—signals a new vector for leadership engagement, particularly for distributed organizations with tens of thousands of employees. By making C-suite perspectives available on demand, businesses can flatten hierarchies and accelerate decision alignment without overwhelming executive bandwidth.
Key Business Value:
Scalability of leadership guidance and knowledge sharing
Potential for AI-driven internal support tools tailored to specific organizational culture and strategy
Enhancement of onboarding, communications, and strategic transparency at scale
OpenAI Codex Update: Mac Automation, Parallel Agents, and User Memory
OpenAI’s latest Codex update (particularly for Mac users) now offers simultaneous agent operation with independent system-level cursor control, seamless app and file management, integrated web browsing, built-in image generation, and persistent workflow memory. This creates a foundation for super-app development—where AI acts as both assistant and executor, automating complex multi-app workflows in the background.
Business users benefit directly through hands-free process automation and proactive workflow suggestions, making knowledge work more leverageable and repeatable. The comparison against other desktop agent platforms (such as Anthropic’s Claude Code) reveals superior speed, consistency, and utility for business operations that depend on fast, reliable multi-app automation.
Key Business Value:
Material reduction in manual work through system-level AI automation
Improved process consistency and knowledge retention via persistent workflow memory
Enhanced ability to pilot AI agents alongside human users on a single device, minimizing workflow interruption
Ancillary AI News: Infrastructure, Staffing, and Market Movements
Additional highlights provide further business context:
Google’s Gemini desktop app for Mac and partnership moves reflect a trend toward platform-agnostic AI integration.
Canva’s prompt-based automation features lower the barrier for marketing and design-dependent organizations.
Sector-wide job restructuring tied to AI adoption delivers clear business signals on where efficiency and headcount decisions are trending.
Shifts by traditional product companies (including pivots into AI compute infrastructure) suggest increased competition and opportunity in the broader technology supply chain.
Key Business Value:
Early identification of platforms and vendors most likely to offer stable, performant integrations
Awareness of sector-wide precedent for restructuring and efficiency-focused deployment
Real-world metrics for assessing vendor sustainability and technical support models
Conclusion: Extracting Immediate Practical Advantage from AI’s Evolving Landscape
The coordinated product releases, internal strategy shifts, and executive decisions detailed above are not simply signals of technological progress—they provide roadmaps for practical business advantage. The granular advances detailed in routines, multi-agent automation, design asset pipelines, metric transparency, and leadership scaling offer actionable upgrades for enterprises and SMBs alike. Decision-makers leveraging these specific insights can identify both immediate operational efficiencies and longer-term pivots that align with AI’s rapid market progression.
Topics Covered in This Episode:
- Anthropic Claude Code Desktop App Upgrades
- Claude Opus 4.7 Model Vision Capabilities
- Anthropic Claude Design Editable Visual Assets
- OpenAI Revenue Dispute vs Anthropic Explained
- OpenAI Shareholder IPO Leadership Concerns
- Salesforce Headless 360 Agentic AI Platform
- NSA Deploys Anthropic Mythos Preview Model
- Meta AI Zuck: CEO Digital Avatar Project
- OpenAI Codex App Super App Preview
- Codex Mac Desktop OS-Level Automation
- Google Gemini Desktop App for Mac Released
- Canva AI 2.0 Prompt-Based Design Tools
- OpenAI GPT-5.4 Cyber for Advanced Defense
- Allbirds Pivot to Newbird AI Compute Infrastructure
Episode Transcript
Jordan Wilson [00:00:16]:
This week in AI updates, we had the wildest twenty four hours of new features that I've seen since covering AI daily for more than three years. I mean, first, we saw Google release their version of Gemini on the desktop for Mac, which could be a major focus point for Google in the future to start competing on the desktop, which is where the competition is clearly headed. Then we saw Aithropic release their Opus 4.7, which will likely and officially reclaim the crown as the top AI model in the world, but not sure for how long they'll have that. Then we saw Perplexity respond by releasing their Perplexity computer agentic desktop, product after announcing it. Well, now it's out in the wild. And then we got what I think is one of the bigger releases we've seen in the last year or more because we got immediately a preview of OpenAI's new super app, at least unofficially, as they unveiled their new codex update. And my gosh, this just created a new tier of computer use products. And right now, OpenAI is in its own class.
Jordan Wilson [00:01:30]:
Yeah. If you weren't paying attention every single day this week, you probably missed a lot of new features that are going to impact how you work. So if you can't spend hours every single day like I do keeping up with it, then you should probably be joining us on Fridays for our new segment that we're calling Friday features. Alright. So here is what you're gonna learn on today's show. You're gonna learn why proactive AI may become the default sooner than later. You're gonna see why a second mouse is the small but big update. The smallest but big update we haven't seen in a while.
Jordan Wilson [00:02:10]:
And then you're gonna see how personal computers might be back in a big way, and I'm not necessarily talking about OpenClaw, though we will give you the OpenClaw updates. Alright. Let's get into it. If you're new here, welcome to Everyday AI. My name is Jordan Wilson, and, well, we do this every day. This is an unedited, unscripted livestream podcast and free daily newsletter helping everyday business leaders like you and me keep up with the AI craziness. I tell you what's new, what matters, how to use it, and how to get ahead and use that information to grow your company and your career. So if you're on that journey like me, like millions of people around the world, awesome.
Jordan Wilson [00:02:46]:
Starts here, but make sure you go to our website at youreverydayai.com. We're gonna be recapping today's show in the newsletter. So if you're out walking your dog or on a treadmill or if your dog is walking you on a treadmill, and you miss something that we talk about or you're like, hey. I need to go use that. It's all gonna be linked in today's newsletter. Alright. So let's look live. Let's talk about some of the big AI features that we have going on in AI world.
Jordan Wilson [00:03:13]:
So, and, hey, I haven't done this in a while, but, you know, livestream audience, if you have any any questions about the tools, you know, we'll go ahead and and answer those. So, yeah, let me know. Haven't haven't done the answering questions in in a while. So if if you have questions as we go along, to those on the live stream, let me know. So first, as I sip my very hot coffee, anthropic. Yeah. We got the by definition, for the most part, the newest and most powerful model in the world today with Opus 4.7. So this replaces Opus 4.6 as the new default best model for anthropic, and they say it is meaningfully better at coding, seeing images, and following instructions precisely.
Jordan Wilson [00:04:06]:
So it can now process images at three times the resolution of its predecessor, and that part is actually very important. Sounds like a minor dorky detail. It's like, okay. Why does that matter if we can, process images better? Well, that means it can output them much better. So we're talking about, even better diagrams, documents, screenshots, etcetera. So, you know, I think a lot of people have been using, Anthropic's products to create things like decks or to create spreadsheets that are well formatted. So now, Opus 4.7 may be even better at that. And then Anthropic says this model obviously is not as good as the best model as it has.
Jordan Wilson [00:04:46]:
Right? So, we still have the unreleased Mythos model, which remains behind, the cybersecurity only program called Project Glasswing. So, who has access? Well, everyone on a paid account. So no matter where you're using, Claude, whether it's claude.ai, the API, Amazon Bedrock, Google Cloud, Microsoft Foundry, wherever, the desktop, you're gonna see Opus 4.7 as the new choice at the top of the list. If you are using it on the API side, good news for that because it is the same price in terms of input output, tokens if you're building on the API. So it's $55,000,000 input, 25,000,000 out, $25 per million outputs. And there is a, on the Claude code side, there's a new feature, called auto mode. Alright? So you can get a new processing intensity setting, for max subscribers. Alright.
Jordan Wilson [00:05:42]:
Here's why it's useful. Well, as with any AI model, right, the better the model itself gets, the better results you're gonna get. Right? So, if I'm being super honest here, this is probably at least for me so far, and I've haven't got to play with it a ton. But, you know, when we went from, four five to four six as an example, I saw a noticeable and measurable difference. At least it's only been a couple of hours. Right? This is a very fresh model. I haven't seen that same, kind of jump so far, between 04/06 and four seven. But, again, I've haven't had enough time, to try, I think, what will truly be, the most useful features, which is better looking slides, decks, etcetera.
Jordan Wilson [00:06:31]:
So who's gonna find this useful? Well, developers using Claude code for complex multi step code coding tasks. So I think that you are gonna see better results on the coding side. Obviously, anyone doing computer use or document analysis that was frustrated by some of the, image inputs, outputs, that's gonna be a little bit better. And then enterprise teams that need Claude to follow detailed instructions precisely, you're gonna see improvements as well. So, here's what Anthropic said. They said Opus 4.7 is a notable improvement on Opus 4.6 in advanced software engineering with particular gains on the most difficult tasks. Users report being able to hand off their hardest coding work, the kind that previously needed close supervision, to Opus 4.7 with confidence. Opus four seven handles complex long running tasks with rigor and consistency, pays precise attention to instructions, and devises ways to verify its own outputs before reporting back.
Jordan Wilson [00:07:26]:
The model has also substantially better vision. It can see images in greater resolution. It's more tasteful and creative when completing professional tasks, producing higher quality interfaces, slides, and docs. And although it is less broadly capable than our most powerful model, Claude and Mythos preview, it shows better results than Opus four six across a range of benchmarks. So at least in some of my early testing, I think maybe I had some of my, kind of saved workflows, maybe over optimized for Opus 4.6. Or one thing that you're gonna notice that you might not like with Opus 4.7, which is interesting because Anthropic didn't even mention this. So it is the first model with adaptive reasoning. Okay? So when you are using the Claude models, at least on the chat side at claude.ai, there is a feature that you can toggle off and on.
Jordan Wilson [00:08:18]:
And that feature is called, extended thinking. Right? I'm I'm actually just double checking to see if they, changed it. And this is strange. Okay. This is why I love doing things live because literally thirty minutes ago, it wasn't like this. Okay. There we go. Little bug on anthropic side.
Jordan Wilson [00:08:39]:
They should probably figure that one out. Alright. So, normally, you would have an extended thinking toggle. Alright. So that forced the model to think a little bit more. So if you're still using Opus 4.6, you can force it to think longer. Right? So on Opus 4.7, this is the first one that you do not have that exact toggle. Instead of the extended, which forces the model to think longer, now you have adaptive thinking, which, again, Infraumpe didn't even talk about this, but it does appear and it's not even on their blog post release, but it does appear that the model will choose, in this case, if it will think harder or not.
Jordan Wilson [00:09:21]:
So some of the few test prompts that I ran, Opus four six versus Opus four seven with, the extended thinking versus adaptive thinking, four six was actually better. And I think because the model thought, oh, for this one, I don't need to think very well. But it is intentionally a trickier prompt, that I use to prepare for actually this show, right, just to verify some some facts and stats and things like that. And, the four seven didn't do as good. So I don't know if it's the adaptive thinking or maybe if some of my saved prompts were a little just too, set up for, four six. Alright. Next, let's get going here. Sorry.
Jordan Wilson [00:10:04]:
Getting all my windows my windows mixed up here. Alright. So do see a couple questions. Lisa says, will the price of tokens for Sonnet drop? Maybe. I would assume right now, probably not because a lot of people are using Sonnet on the API side for things like OpenClaw. So I would assume that Infrapping is not going to drop the prices for any of their four six models aside from Opus four, four six if they do. I I'm guessing they probably won't. Alright.
Jordan Wilson [00:10:35]:
Jackie here said, Claude helped me iterate on a presentation I gave yesterday. This morning, I went to high five Claude as we killed it. Yeah. One thing that stinks, you know, about entropic in general, and they're the only one of the big three that doesn't do this anymore, is if you start so all your old chats that you were working on with maybe Opus four six or Sonnet four six and you're like, oh, I wanna see how, you know, Opus four seven does on this. Well, you gotta start over. You can't, you know, change your model during the chat. So but if you're using, you know, skills, projects, etcetera, it won't matter as much. But if you have an older kind of project that you were working on inside of a chat, you can't switch over.
Jordan Wilson [00:11:13]:
Alright. Our next AI feature update and the rest are gonna be a little quicker. So we had a Claude code desktop redesign and the release of routines. Alright. And FYI, I did go over this actually Wednesday in episode seven fifty six, so this one's gonna be a little quicker. But Infrafx completely rebuild the Claude desktop app with a lot of those new features coming in Claude code for Mac and Windows. The big thing is there's a new layout for managing multiple coding sessions at once, including a sidebar showing all active and recent sessions in a shortcut to ask side questions without derailing a running task. That part is really cool.
Jordan Wilson [00:11:55]:
So if you've used cursor, right, you might be familiar with that kind of setup. It is cursor esque. Right? But, they also launched, which I think is the bigger news than just the, you know, the nicer looking desktop app and some new features in quad code. The big thing here that I don't think got enough love is the routines. These are essentially saved automations, a huge improvements on kind of, Cloud Code's previous res, scheduled tasks inside of Cloud Code. So, essentially, in this, you define a prompt, connect it to your code repos, and run it on a schedule, or on triggers, and that's the big thing. So you can set up a, you can set up a local routine or you can set up a cloud routine. So little differences, obviously, with a local routine, you can, access everything on your docu like, on your desktop, and you can have it use your browser as well.
Jordan Wilson [00:12:55]:
For remote routines, a little bit different. So you could essentially have them kick off once something happens. So, you know, the big thing here that, again, people aren't really talking about is, you can have it set off with a trigger like an API call. So, essentially, it's a web hook. Right? If you know anything about marketing automation, you you know, so now all these different things on the web, right, or anything you connect to Zapier can happen, and then that can schedule a, routine inside of Claude. Right? So anything that you would normally do inside of claude.ai. Right? Think outside of this being inside of Claude Code. Right? The downside is this lives inside of Claude Code, so I think nontechnical users are gonna be scared of it, But it's literally the simplest thing.
Jordan Wilson [00:13:37]:
I went over this, on Wednesday's show. So that's the big thing that literally anything now can happen on the Internet. And then you can have a cloud routine that automatically runs anything that you would normally run. Alright. This is a huge unlock for proactive AI versus reactive AI. So who has access to this right now? So the new desktop design has rolled out to all paid users, and here's why it's useful and who I think is gonna find it valuable. So instead of working on one thing at a time, right now developers can use, kind of the new multi agent orchestration features to get a lot done, at the same time. And then routines are huge.
Jordan Wilson [00:14:16]:
Right? That can automate the repetitive stuff for literally anyone. Right? Whether you're, you know, doing daily code reviews or something I'm testing it out for is, you know, grabbing a lot of my stats every single day. Things that I would normally manually do, go in and see, okay. How did yesterday's podcast do? How did yesterday's newsletter do? You know, how did all these things from yesterday do at everyday AI? A lot of times would take me a lot of clicking and thinking. Right now, I can get a great draft version of that with doing absolutely nothing first. So this is, like I said, a big jump toward proactive AI. Alright. Oh, I forgot to talk about this, the, the new model updates, but, benchmarks, we shared that in our community.
Jordan Wilson [00:14:57]:
If you wanted to see how Opus 4.7 stacked up against Meta's, Muse Spark, Gemini three one Pro, and OpenAI's GPT five four in Mythos preview. We did put that in our, community. So, next, what's new here? Big update, I think. Perplexity's personal computer. Yeah. Trying to redefine the personal computer. So here's what's new. Perplexity computer takes cloud based AI agents and brings it directly onto your Mac, giving it access to your local files, native apps, and browser.
Jordan Wilson [00:15:36]:
So you can press both command keys on your computer to activate it, and then it can respond to either text or voice. It can see what apps you have open and suggest quick actions immediately. So it was announced originally in March with a wait list, but then any max subscribers are automatically get access to it, and anyone that was on the wait list gets access to it as well. So now for the first time, this one is officially live. And to put this simply, this is Perplexity's version of Open Claw. Right. So here's, you do this is Mac only right now, FYI, and it does require the latest version of Mac, and there's no Windows version announced yet. So here's why it's useful.
Jordan Wilson [00:16:25]:
Well, it just works entirely across your Mac. This is one of those where, like OpenCloud, to do it best, you're probably gonna wanna have it, work on its own dedicated computer. Alright? Because it will be able to read and write local files, open apps, browse the web, connect to cloud tools, all from a single conversational interface. So in the demo that, Perplexity put out there, essentially, someone had a a list of tasks open on their computer, I think, in Apple Notes. It just said, hey. You know, via voice, go do my tasks. Right? And it went through, pulled up all the right, pulled up all the right apps. I think it, you know, created, you you know, created a calendar event, some of the basic things.
Jordan Wilson [00:17:04]:
So, you know, give it a try. Right? If you have wanted to try OpenClaw as an example, and it looks a little bit too technical or maybe you're not, you know, on board with some of the safety ramifications of OpenClaw, you want something a little more sandbox, I think personal computer is a great step. Right? Things like Claude Code, codex, which we're gonna be talking about a little bit more here. You have a little bit more control, and it's not a truly kind of always on autonomous agent. Although with Cron Jobs, right, which is essentially just scheduled AI agent runs that can happen, either on a schedule or, on a trigger. It essentially turns it into that. Right? Whereas things like personal computer from perplexity or OpenClaw, you know, it it is much more, goal based versus task or project based. So that's the kind of big differentiator.
Jordan Wilson [00:17:58]:
So here's, what perplexity said in their release. They said, today, we're beginning the rollout of personal computer. Personal computer is a powerful expansion of perplexity computer. We actually did a perplexity computer show about a month or so ago. It's actually really freaking good. So I I I will I mean, we'll see if I have time to test personal computer. I should have time in the next, I don't know, month or so. Too many updates, right, to test all these things all individually aside from very topical tests.
Jordan Wilson [00:18:27]:
So it says perplexity computer brings the multi model orchestration of computer to your machine. It can work across your local files, native applications, connection, connectors, and the web to complete complex and even continuous workflows. Personal computer makes perplexity computer a more personal orchestrator, eloquently hybridizing the local and server environments for maximum security and productivity. So when I did my original perplexity computer, not personal computer episode, I was actually really freaking impressed because it's really good. And here's what is different and unique, about computer from perplexity both on the web and on the desktop. It uses kind of that model orchestrator. So it has access to all of these different models. So even if you don't subscribe to Claude, to Gemini, to OpenAI, It's gonna use the best model for the right job automatically, which sounds absolutely amazing.
Jordan Wilson [00:19:27]:
Right? The huge downside is this burns through credits. Burns. Right? I was on the $200 a month plan. I did one single task when this first rolled out, and it ate up 10% of the weekly allowance on a $200 a month plan. Yes. It produced something that was very useful, but eats the credits. So who's gonna find this valuable? Well, like I said, anyone that is, wants to kind of get into setting up a computer as an always on agent and apparently just has unlimited budgets because that's the downside. It will eat through, pretty quickly.
Jordan Wilson [00:20:05]:
But anyone that's privacy conscious, or if you just have a Mac mini, right, and you really wanna turn that machine into an autonomous AI workflow, there you go. But right now, if you really wanna get any benefit of it, you're gonna have to be on that $200 a month plan, for the max, and it is still fairly expensive. The good thing is it does include probably a little stricter safety guardrails, and user approval is required for sensitive actions. Alright. Yeah. Jackie here says OpenClaw does intimidate me. OpenClaw is great. I think some of the, more recent updates over the last couple of weeks have made it a little bit, more secure and a little bit more, consistent.
Jordan Wilson [00:20:48]:
And we're gonna talk at the very end of the show some of the big open claw updates for the past week, but a lot of them have been on, performance. Alright. Let's move on to Google is coming to the desktop. So finally, Google released a standalone Gemini app for Mac. So they previously had a Gemini app for Windows. Now Google Gemini is available for Mac, and it's built natively in Swift. So it lives on your desktop instead of in a browser tab. So, essentially, you can share any window on your screen with Gemini so it can see what you're working on and answer questions in context.
Jordan Wilson [00:21:28]:
So here's the thing. Who has access? Well, it's free for anyone to download and use. You obviously connect it to, your Google Gemini, plan. So it's free to use. Right? But it's not like you get new features by using this. It is still tied to the features that are available on your plan. So you do have to have a newer Mac. It does require an Apple Silicon Mac.
Jordan Wilson [00:21:51]:
So as long as you bought your Mac in the last I don't know what that's like four years, you should be fine. But if you have a real old dusty Mac, this isn't gonna work there. So here's why it's useful. Well, no more opening a browser tab, to talk to Gemini. It's now a keyboard shortcut away, from using anything on your Mac, and that's the good thing. The the the big thing that's makes this more useful, than using Gemini in the browser is you can instantly share your window because that gives Gemini instant visual context. So if you're working on a chart or if you're trying to share the window and say what are the three biggest takeaways of this presentation, etcetera, it just makes it easier without having to copy and paste everything over to Gemini or without having to do a screenshot. Right? So you can just instantly bring the context in.
Jordan Wilson [00:22:38]:
So here's what Google says. Right? Kind of the big highlights. They say it's a shortcut to Gemini with a simple global shortcut. Gemini is instantly ready, when you are to tackle any project, being able to give it contextual help. Right? So based on any of the documents that you're using, so you can pretty quickly, just give Gemini visual access to any other window on your computer as well as you can still upload files. And it obviously brings a level of creativity, you know, to your desktop because anything, you know, whether you're creating images, video, music, right, with a new Lyria three, model, you can now do all of those things on your desktop. So who's gonna find this useful? Well, I if you're a Gemini power user, and you find it sometimes a waste to have to do a lot of copying and pasting, or sharing screenshots to get Gemini to do something, you're obviously going to love this. Google does call this the first step toward a fully proactive desktop assistant, and, obviously, we're gonna get be getting more updates on that next month at Google's big IO conference.
Jordan Wilson [00:23:45]:
Alright. So what do you guys think? Are are like, is everyone out there are you using at least for me, I'm using browser AI way less. I will say probably now, between, you know, Claude code, Claude co work, the Claude desktop app in codex. That's probably, like, 60% of my use now. You you know, being working on the desktop, it does just have a whole new unlock. Right? Especially with the new routines, from Anthropic and what you can kind of schedule, now with, codecs, which we're gonna be talking about here in a minute. Alright. Our next one, a small one, but pretty cool if you like free AI.
Jordan Wilson [00:24:35]:
Alright. That's because Meta started rolling out contemplating mode for Muse Spark. So we covered Muse Spark last week, which is Meta's new model. Alright. This is not open source unlike their llama models, but right now, it is a 100% free to use. So Muse Spark as a model itself actually did fairly well on benchmarks, which, you know, some people were kind of surprised by. But given how much Meta, invested, like, billions of dollars and paying, individual researchers, reportedly, like, a $100,000,000, it makes sense that their model was actually pretty good for the first version, of Muse, and Meta did say that they're gonna have, future, variations of the Muse model. I'm guessing it'll be Muse whatever.
Jordan Wilson [00:25:25]:
Right? Not Muse Spark, but Spark is supposed to be the smallest one. But what is new this week that you may have missed is they have a new contemplating mode. So here's what Meta says. They says, we're also releasing contemplating mode, which orchestrates multiple agents that reason in parallel. This allows Muse Spark to compete with the extreme reasoning modes of Frontier models, such as Gemini Deep Think and GPT Pro. Contemplating mode provides significant capability improvements in challenging tasks, achieving 58% in humanities last exam and 38% in frontier science research. So, like I said, the reason why I think this is important to talk about and why I think people will find this useful is those other modes there that I just talked about, Gemini 3.1 Deepthink and GPT five four Pro, you essentially have to be on a $200 a month plan to get that level of intelligence. That is like the creme de creme of AI models.
Jordan Wilson [00:26:26]:
Now you can get it for free. So although I think Muse Spark by itself, yes, it's good. It's it's a very capable multimodal, multimodal model. But to be able to go use the contemplating is really cool. So how this works, it's 16 agents that are working in parallel. So it does break these complex tasks tasks down, and it is kind of cool to watch how it breaks the task, or the bigger project into smaller tasks. So like I said, anyone that needs a strong reasoning model or maybe if you, are only on free plans, which I would never recommend. Right? Most of these plans, their baseline are $20 a month, and you should always, always, always be using thinking models.
Jordan Wilson [00:27:10]:
But if for whatever reason, you just can't pay for AI, it's at least going to work, worth checking out not just MuseSpark, but the new contemplating model that was not available at launch, but they did slowly roll it out last week. Alright. Let's get going to skills. Right? We got more skills this time inside Google Chrome of all places. Google Chrome. Alright. Here's what's new that you may have missed. Google added a new skills feature to Gemini in Chrome on desktop.
Jordan Wilson [00:27:48]:
So a skill overly simplified essentially is a saved prompt that you can reuse with one click on any web page across multiple tabs. And that's the cool thing here about the new Gemini in, or sorry. Yeah. The new Gemini in Chrome. Alright. You can use that across multiple tabs. So essentially, here's how it works. After running a prompt in Gemini Chrome's side panel, you can save it as a skill.
Jordan Wilson [00:28:12]:
So the next time you can type a backslash, and then you can see all the new skills that you have saved, and it will run on whatever page or pages that you're viewing. And then skills are based, or are synced to your Chrome account, which is cool because if you're someone like me that uses multiple computers, that's gonna sync and then you can use those skills across any of those. So this is rolling out to anyone on Chrome desktop, who has set their Chrome language to The US. This is free. It does not even require a paid Google subscription, and that's why I think this is huge. So if you haven't used the, Gemini sidebar on Chrome, you can even see it, kind of I'm sharing my screen here on the live stream. It's just in the upper right hand corner. So you just click that, you know, and then you can ask and chat, with Gemini on any page or pages or tabs that you have open.
Jordan Wilson [00:29:12]:
And now you can save all of those, kind of prompts as skills to reuse them later at any time. Here's what Google says. They said until now, repeating an AI task, like asking for an ingredient substitution or to make a, you know, to make a recipe vegan, you you know, that meant reentering the same prompt as you've been visited different pages. To make this easier, we're launching skills in Chrome, which lets you save and reuse your most helpful AI prompts and run them with a single click. When you write a prompt that you'll want to use again, you can save it as a skill directly from your chat history. The next time you need it, select your saved skill in Gemini Chrome by typing forward slash, clicking the plus button, and then your skill, you know, then you select your skill, and then your skill will run on the page you're viewing along with any other tabs you select. You can edit your saved skills and create new ones at any time. So a great quality of life that I think anyone will be able to use.
Jordan Wilson [00:30:12]:
This is one of those even just Gemini and Chrome in general, that even if you don't have a paid plan, this is just a great way to more quickly browse the web. For me personally, I spend way less time browsing actual websites than I did two years ago. Right? But when I do browse actual websites, if I'm not using OpenAI's Atlas browser or if I'm not using Perplexity's Comet browser, then I'm obviously using Chrome, and I am using the Gemini sidebar a ton. Alright. Our last big updates of the week and y'all, this one, absolute banger. Codex, y'all, since codecs got released as an app, I've been the weird guy saying, this is better than Claude Cowork. This is better than Claude Co. I think a lot of people saw the word code in codex and ran for the hills, but we got, a ton of new updates in codex, and this is the first step toward the super app.
Jordan Wilson [00:31:17]:
Alright. So let me first give context on what that means. So OpenAI right now has three different desktop apps. They have the codecs app, which is probably a little bit geared more toward developers. They have the chat g v t app, alright, which is your normal chat g v t. And then they have the Atlas app, which is their agentic browser. So, OpenAI has announced that they're going to essentially be focusing on just one app in the future, and that is going to be their super app. We don't know what the name is gonna be.
Jordan Wilson [00:31:48]:
But with this update, yesterday, OpenAI did say this is the beginning of their super app. So codecs the codecs lead in a media briefing did confirm this, and they said, we're actually doing the sneaky thing where we're building the super app out in the open and evolving it out of codecs. So I've been telling you guys, and I'm gonna I'm not gonna stop screaming because codex, it is especially with this update, it's better than Claude Co work. It's better than Claude Co, and it is not even close, especially with this new computer use update, which I'm gonna confirm is why it's such a big deal. But let's talk about some of the details. So massive update to OpenAI's codex tool. The headline codex can now control your Mac desktop. It has computer use that is literally in its own category.
Jordan Wilson [00:32:39]:
It can open apps, click, and type with its own cursor. That's the big that's the headline here y'all. I've talked about how, yes, I love, Claude's computer control. But one thing you've heard me talk about over the last couple of months is it screen jacks, it cursor jacks, it type jacks. Right? So what that means is the only real way that you can use computer use if you're using Claude code, or it works on Claude Co worker as well, is you can't be using your computer because you can't do anything else except sit there and watch it, and it's really slow. So, obviously, what this means to really get the true value out of Claude's computer use, you only should be running it when you're not using your computer, right, which is overnight, essentially. At least for people like me, I'm in front of my computer all day. This is the step change here with codecs in their computer use.
Jordan Wilson [00:33:32]:
I watched this, like, literally a minute after they announced it, and I'm like, wait. How is this working? It literally has its own cursor. So I'm over here taking notes, you you know, as I tried this out, yesterday and chatting with some of the codecs leads on Twitter about it, and they did confirm. I'm like, wait. So this operates at the operating system level and not the user interface level, which it's the only computer use product that does that now. So you can see it working on the interface, but it's actually working in the operating system. So it has its own, it has its own cursor or cursors because you can run multiple of them at the same time. And you can be typing and, having your own window or your own programs up, and it can still work around those.
Jordan Wilson [00:34:21]:
So this is freaking huge. So there's aside from that, which I'm gonna talk about a little bit more here, it also gained built in image generator. Alright. So using g p t five one point five image, it also has a new integrated web browser inside of the app. It has a integrated file browser inside of the app. So a lot of these things that would normally be much slower for a computer using agent to do is now much faster because it lives inside of the codecs app, and it's not pulling all of that context, you know, from, you know, your actual desktop. It's just much faster. Alright.
Jordan Wilson [00:34:59]:
So, here's availability. It's available right now on Mac and Windows desktop, but some of these features are not available yet for Windows. Alright. So there's also a new, memory feature. It has persistent memory, and it remembers how you work across sessions and the ability to even schedule future work for itself based on what you're doing. My gosh. Talk about a proactive AI agent that literally works how a human works, but it doesn't interrupt you. This is huge.
Jordan Wilson [00:35:35]:
So this is why it's useful. It it means codecs can test your apps, click through UI flows, and catch visual bugs without you even touching any things. It it it runs multiple agents in parallel without interfering with your own work. So, yeah, unlike, computer use from Claude that runs on their desktop, it doesn't screen jack. It doesn't mouse jack. It doesn't window jack. Right? It doesn't take over all those things because you can't work at the same time as Claude's computer using agent. Also, the new memory eliminates the need to have to tell codecs how you do the work every single time.
Jordan Wilson [00:36:09]:
It learns your preferences, tech stack, and recurring workflows. And then it also will suggest things that it can automate for you. And then the scheduling lets codecs pick up work across days or weeks. So the same thing how you can schedule routines, inside of Claude Code, on the desktop version on Mac. You can obviously do the same thing in codecs. So who's gonna find this valuable? Literally, anyone anyone. Well, I think developers, obviously, who are using codecs, you're gonna see it right away. But y'all, I I I was not wrong about this when I said in February when codecs was released and no one was paying attention to it, and everyone's like, oh, Claude Coerc, Claude Coerc, Claude Coerc.
Jordan Wilson [00:36:50]:
I'm telling you, codecs right now is where it's at. That doesn't mean I'm gonna stop using Claude Coerc. I love it. I'm not gonna stop using Claude Coerc. I love it. I'm gonna stop using their computer agent because it's slow and I can't work with it at the same time. To me, codecs is better. I think if you need something that looks good right from a design perspective, obviously, the Claude infrastructure is is better.
Jordan Wilson [00:37:12]:
If you need something quickly done, Claude is better. If you need something done right, it's it's it's codex in GBT all the way. Alright. So that is, and let me actually just read a little bit on what OpenAI says. They said codex can now operate your computer alongside you, work with more of the tools and apps you use every day, generate images, remember your preferences, learn from previous actions, and take an ongoing take on ongoing and repeatable work. The codecs app now includes deeper support for developer workflows, like reviewing PRs, viewing multiple files and terminals, connecting to remote dev boxes via SSH, and an in app browser to make it faster to iterate on front end designs, apps, and games. And then they said extending codex beyond coding with background computer use. Codex can now use all of the apps on your computer by seeing, clicking, and typing with its own cursor.
Jordan Wilson [00:38:09]:
Multiple agents can work on your Mac in parallel without interfering, with your own work in other apps. For developers, this is helpful for iterating on front end changes, testing apps, and working in apps that don't expose the API. Alright. So pretty big news. Let me know, livestream people, people listening on the podcast. What do you want more? Do you want more codex, or do you want more Claude code, Claude co work? Let me know. I have my own preferences, but I work for you. Alright.
Jordan Wilson [00:38:37]:
And real quick, OpenClaw updates. Alright. Just quickly here. This week, mainly focused on stability and performance as opposed to new features. They do already obviously have Opus 4.7 reports. Sub agents, get stuck, fewer times. There was big polish, drop for stability and an active memory plug in. Alright.
Jordan Wilson [00:38:58]:
So that is a wrap for the seven new AI features that you need to be paying attention to. So from the new big model, Opus 4.7, Infropics, new routines, which are very helpful. But the banger, in my opinion, on everything, yeah, personal computer might be good, but y'all go check out codex and their new computer using agents that you can work in parallel. Alright. So if this was helpful, please, if you haven't already, go to youreverydayai.com. Sign up for the free daily newsletter. Thanks for tuning in. Hope to see you back tomorrow in everyday for more everyday AI.
Jordan Wilson [00:39:32]:
Thanks y'all.