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
Anthropic, OpenAI, and the Pentagon: What the Latest AI Drama Means for Government and Enterprise
The latest developments in artificial intelligence are anything but business as usual. Recent events in the AI sector—including an unprecedented Pentagon designation, legal disputes, new generative model launches, and major platform integrations—offer crucial implications for corporate strategy, technology adoption, and talent management. Here’s what business leaders need to know based on this week’s most consequential AI news.
AI Supply Chain Risk: Government Actions Disrupt Enterprise Relationships
The US Department of Defense’s official designation of Anthropic and its Claude models as a national security supply chain risk is a new lever in federal technology oversight with practical business implications. Federal agencies and defense contractors are now banned from using Anthropic’s products, and contractors have six months to transition off these tools. Anthropic’s leadership responded with legal action, contesting the applicability of the supply chain risk statute to US companies and framing the policy as overbroad and intended for foreign adversaries—not domestic firms.
For existing Anthropic clients outside direct Department of Defense contracts, reassurance came quickly: Microsoft, Google, and Amazon Web Services all clarified they would continue offering Claude models to enterprise and government clients not tied to defense contracts. This response highlights the importance for businesses engaged with government contracts to understand the changing compliance landscape and to recognize the legal and operational risks when working with emerging AI vendors.
AI Benchmark Competition: OpenAI’s GPT-5.4 Pro and Performance Differentiation
OpenAI’s rollout of GPT-5.4 Pro delivers an 82% win or tie rate against human experts on industry-standard tasks—a direct performance message to both Anthropic’s Claude Opus and Google’s Gemini Pro models. Paid subscribers can leverage this improved model for advanced workflow automation, thanks to a 1,000,000 token context window and a 33% reduction in hallucinations, which translates to greater reliability for business-critical applications like spreadsheet automation, document creation, and presentations.
On the other hand, free ChatGPT users receive access to GPT-5.3 Instant, but not the newest Pro-level features. The rapid iteration among top AI companies means CTOs and operators should avoid workflow disruption by constantly switching “best-of-breed” models; instead, they should build around the capabilities and integration points of their AI operating platform of choice, knowing each leader's advantage may only last a few weeks.
AI Embedded Productivity: ChatGPT and Excel Integration
The new ChatGPT for Excel add-in quietly doubles down on making generative AI an integral part of professional workflows. With direct embedding into Excel workbooks, users now build, analyze, and update complex models with plain language rather than manual formulas. This results in not only speedier financial modeling, scenario analysis, and error tracing, but also seamless inclusion of real-time financial data streams from trusted providers (FactSet, Dow Jones, S&P Global, and more).
This enhancement expands adoption from technical users to frontline analysts, making AI’s advanced reasoning capabilities widely accessible in day-to-day business analysis. For organizations still using legacy Excel macros or custom scripts, the shift to embedded AI tools offers immediate boosts in auditability, agility, and cross-team collaboration—key for faster decision making.
Next-Gen Generative Models: Google Gemini 3.1 Flashlight for Real-Time Operations
Google’s release of Gemini 3.1 Flashlight sets a new bar for speed and cost-efficiency within enterprise-grade AI. Outperforming previous models with a 2.5x faster “time to first token,” and a cost of only $0.25 per million input tokens, this architecture is tailored for high-volume, instant-response applications such as customer support, moderation, and dynamic interface generation. The introduction of adjustable “thinking levels” means technology leaders can trade off depth and speed based on individual workload requirements—an advantage for companies running a mix of rapid classification and more complex code generation.
Autonomous Agents and Open-Source AI: The Business Impact of OpenClaw
NVIDIA CEO Jensen Huang recently labeled the open-source OpenClaw project as “probably the single most important release of software ever” due to its unprecedented uptake and GPU demand. OpenClaw’s agents, which differ from standard chatbots by autonomously completing complex tasks and workflows, now drive previously unseen consumption levels—from 1,000 to 1,000,000 times more tokens per task. For companies investing in AI infrastructure, this signals a hardware and cloud scaling challenge—and a clear opportunity: If deploying autonomous agents, businesses need both robust token budgets and flexible compute strategies.
OpenClaw’s accelerated adoption—surpassing what Linux achieved in decades—highlights a power shift. Major players from OpenAI to Perplexity and Anthropic are building more autonomous features into their ecosystems, reinforcing the trend that future business value will stem not from prompt-based AI, but from agents executing workflows end-to-end.
AI Job Automation Study: Anthropic’s Research on White-Collar Exposure
Anthropic’s recent study identifies a looming disparity: While current AI tools like Claude are technically capable of completing 94% of tasks in computer/mathematics professions, actual enterprise usage hovers around one-third of those tasks. The research warns that professional services—software developers, lawyers, analysts—face outsized automation risk, with those most exposed being higher-income, graduate-educated roles.
Business leaders should pay attention to the concept of “observed exposure,” which tracks not just what AI models can do, but how much is actively adopted in the workplace. As technical and legal adoption barriers fall, the gap will close, and ongoing hiring slowdowns in exposed fields may become more pronounced. Strategic workforce planning now requires reskilling and adaptive task allocation, as the study forecasts a “great recession” scenario for white-collar professionals should adoption rates accelerate.
Key Takeaways for Business AI Strategy
Compliance Risks: Firms with government contracts must audit and adjust tool usage in response to evolving federal AI regulations.
Model Selection: Consistency with an AI operating platform will produce greater value than constantly chasing marginal benchmark improvements.
Workflow Automation: Embedded AI in core productivity tools (like Excel) can double modeling speeds, improve audit trails, and empower non-technical users.
Cost Management: Selecting models based on workload-specific needs (speed vs. reasoning depth) can dramatically lower operating costs, especially in real-time use cases.
AI Infrastructure: Autonomous agents will drive unprecedented compute requirements and may redefine how companies budget for cloud and hardware resources.
Talent Strategy: As AI’s technical abilities outstrip adoption, leaders should focus on reskilling and role redefinition in at-risk white-collar domains.
These specific shifts in the AI landscape aren’t hypotheticals—they’re already impacting enterprise contracts, product selection, and staffing. Staying current on the latest actionable changes shapes a forward-leaning AI strategy that moves beyond pilot projects and delivers quantifiable results.
Topics Covered in This Episode:
- Anthropic vs Pentagon Supply Chain Drama
- Pentagon Bans Claude AI for Defense Use
- OpenAI Launches GPT-5.4 and GPT-5.4 Pro
- GPT-5.4 Industry Benchmark Performance
- ChatGPT for Excel Beta Integration
- Google Gemini 3.1 Flashlight Model Release
- Jensen Huang Praises OpenClaw Agents
- OpenAI Developing GitHub Alternative
- Anthropic Study: AI White Collar Job Disruption
- Latest AI Feature Updates: Claude, Copilot, Gemini
Episode Transcript
Jordan Wilson [00:00:15]:
This week, the ongoing anthropic versus the US government saga continued to draw the most attention in AI world. But in between the supply chain risk designation and anthropic saying they're suing the government. OpenAI and Google both dropped pretty consequential models in their own rights. Then NVIDIA CEO Jensen Huang unexpectedly showered one open source project with the highest praise possible while your spreadsheets just got a new friend that will do all the work for you. Speaking of doing all the work for you, Anthropic came out with a report that basically said, well, here's the jobs our AI models are going to do for you. There was a ton of news in AI this week, and if you missed any of it, if you don't have hours a day to keep up and be like, what does this mean? Don't worry. That's exactly what I do on Mondays on Everyday AI. Welcome.
Jordan Wilson [00:01:15]:
What's going on y'all? Welcome to Everyday AI. My name is Jordan Wilson, and this is for you. It's your daily livestream podcast and free daily newsletter helping everyday business leaders like you and me keep up with the breathtaking pace of AI, help you keep track of what matters, what doesn't, and how to use it to grow your company and your career. So if that's what you're trying to do, starts here with the unedited, unscripted livestream podcast. But if you wanna be the smartest person in AI at your company, our website, that's where you make it happen. Youreverydayai.com. Go sign up for the free daily newsletter where each day we recap that day's, podcast as well as everything else that you need to know. Alright.
Jordan Wilson [00:01:56]:
Let's get into it. On Mondays, we bring you the AI news that matters. So if you don't have hours every single day, well, Mondays is where it's at. Let's get straight into it. The big AI news story of this week was undoubtedly the ongoing anthropic versus the US government saga, and it did not disappoint. If you like drama, if you like to see what's happening under the covers, we got a lot of exclusive reporting this week. So the very public beef between the US government and anthropic continued after the Pentagon late last week officially designated Anthropic as a national security supply chain risk, a move that could reshape the relationship between the tech firm and the US government. So on Wednesday, the Department of Defense officially labeled anthropic and its AI products a national security supply chain risk immediately banning federal use and threatening the company's access to government products.
Jordan Wilson [00:02:59]:
So the Department of Defense said, hey. From here on out, you can't use anthropic clod models. And they actually said, you have six months to transition, but we've seen reports that they're still using it. So Anthropic's CEO, Dario Amati, in response announced that the company will sue the US government, arguing that that designation is legally unsound and meant for foreign adversaries, not US firms. So the conflict began as we've been covering it over the last few weeks when Anthropic reportedly refused Pentagon demands to remove certain wording around safety guardrails that blocked its AI from being used for fully autonomous weapons in mass domestic surveillance. So the Pentagon insisted on, quote, unquote, all lawful use of Anthropic's AI, but Anthropic held firm on what they said were its ethical restrictions leading to a breakdown in negotiations. And that's not even all of the drama. That's because this week related to this, there was also a leaked internal memo that came out from Dario Matti that set off a firestorm after he criticized OpenAI for quickly agreeing to the Pentagon terms, calling their approach safety theater and accusing them of prioritizing employee appeasement over meaningful safeguards.
Jordan Wilson [00:04:22]:
Amodi later apologized for the tone of his message, but reaffirmed what he said was Entropic's commitment to safety and responsible AI use even in military partnerships. OpenAI, meanwhile, signed its own deal with the Pentagon last week and revised it, moving quickly to fill the gap left by anthropics ban. The new supply chain risk designation is pretty big because, according to reports, this is the first time it's ever been used in this way against a US company. And it could force all defense contractors to stop using Anthropic's clawed model if they are working with, Anthropic directly. Though, Anthropic argues the legal reach is narrower than the administration claims. So, yeah, there's a lot of, you know, kind of he said, she said when it comes to the application of this supply chain risk. So in Propic is essentially saying, yeah, this doesn't really number one, this doesn't apply to us, and we're gonna sue the US government, and this is not lawful. And the, the government is saying, well, nope.
Jordan Wilson [00:05:28]:
You know, if you do business with the government, you can't use Claude's models. And they said no more immediately, and they're giving, different departments six months to find a plan. So Anthropics lawsuit will challenge the use of the supply chain risk statute against a domestic company, first amendment. So I have no clue what's gonna happen here. Reading it myself, I'm like, okay. I see where anthropic is coming from. It does seem like a little bit, of a broader application of the supply chain risk than what it was originally meant for. But I guess it's up to the current administration to decide how they want to try and apply certain laws, and I guess that's why we have the judicial system, to begin with.
Jordan Wilson [00:06:14]:
So I do expect that this is not going to die down, and, actually, our next news story does relate to it. So let's get straight into that. So the Pentagon's ban on Anthropic's Claude AI for defense contracts has sparked concern, but tech giants say that most users won't feel the impact. That's because Microsoft, Google, and Amazon Web Services have confirmed that Claude AI remains available to all of their customers except those directly tied specifically to Department of Defense contracts. So Microsoft said they will continue offering Anthropics models through m three sixty five, GitHub, and AI Foundry for enterprises, syrups, and general business users. Google announced Claude remains accessible on Google Cloud for non defense workloads with no changes expected for those other, client. Then AWS told CNBC that its customers and partners can keep using Claude for non defense projects. So these assurances mean that businesses and professionals using Claude via Microsoft, Google, or AWS, even if they are in government roles, do not do not need to worry about losing access.
Jordan Wilson [00:07:31]:
So the Pentagon designation, sorry, the Pentagon designated Anthropic as the supply chain risk after, well, the company refused to provide unrestricted access to its AI for military usage. So, just as a kind of reassurance to everyone out there, because I've even seen a lot of comments from our audience wondering if they work in the government and they use Claude models. So, at least right now, the way that the big tech and their partners, right, because, Google, Microsoft, and Amazon are all investors in Anthropic and they provide Anthropic's clawed models to their, clients. Right? Which include a lot of big government, agencies. So right now, the way that even those big partners, Microsoft, Google, and Amazon are, kind of framing this is that the restriction only replies to defense related contracts and direct contractors and other government, kind of, clients are not impacted. So yeah. Like I see, we'll see this one. I'm sure it's gonna shake out in the courts.
Jordan Wilson [00:08:37]:
It's probably going to take a while. My hunch is this is probably not going to impact a whole lot of people unless you actually work in the department of defense. But, again, that's why I'm not a lawyer, and I just talk AI every day. Alright. Speaking of talking AI every day, I talked about this one late last week because it was big. Our next piece of AI news, OpenAI has launched its latest frontier model, GPT five four thinking and GPT five four pro, making a major leap in capabilities and positioning itself squarely more so against competitors, Anthropic and Google. So, here's what you need to know, and we did do a deeper dive on Friday. That's episode seven twenty eight, so make sure you go check that one out.
Jordan Wilson [00:09:25]:
So OpenAI's GPD five four Pro now achieves an 82% win or tie rate against human experts in the GDP valve benchmark, meaning it matches or outperforms industry professionals in real world tasks 82% of the time, which when you think about it is absolutely bonkers. So the new models are already available immediately to paid subscribers across ChatGPT, the API, and the codex platform. But right now, free users, not so much. That's because OpenAI also had a kind of confusing rollout last week because, well, free users have a new model to use as well, and that's GPT five three instant. So that is because OpenAI released that, I think, the day before they released GPT five four. So, free users and the new default model even for paid users is GPT five three instant. So, yeah, a little confusing. We went from, this point last week being on GPT five two to then they announced GPT five three instance to then they immediately skipped over, you know, there's no GPT five three thinking, no GPT five three pro.
Jordan Wilson [00:10:40]:
We went straight to GPT five four thinking and pro. Yeah. Little confusing. But what's not confusing is the benchmarks. So, on most scientific benchmarks, GPT five four is outperforming Infropix Claude Opus four six and Google Gemini three one Pro. In most categories, even those related to computer use, tool calling, and real world tasks, which is generally where Anthropics and Google's models excel. So the other big thing to keep an eye on is 1,000,000. That's because in the API and on codex, g p t five four features a 1,000,000 token context window for long running tasks and projects.
Jordan Wilson [00:11:23]:
So OpenAI says the model has significantly reduced hallucinations as well, cutting that down by 33%, making it more reliable for business critical work such as spreadsheets, presentations, and document creation. So OpenAI has emphasized native computer use and tool integration, allowing the model to operate desktop and browser tasks, use tools more efficiently, and interact with a user's machine for complex workflows. So I know this it's it's it's like this cycle. Right? Because, essentially, the last three and a half weeks, we've gotten models from all the big players. Right? We had well, kind of in consecutive weeks, OpenAI or sorry. Anthropic had their, Ocus four six followed by SONNET four six, then we got Gemini 3.1 Pro, and now we have OpenAI GPT five four. So, you know, it's it's one of these things. Each time for the most part, one of these three companies releases a new point, model.
Jordan Wilson [00:12:21]:
Right? So going from Opus four five to Opus four six, you know, Gemini three to Gemini three one Pro, same thing with OpenAI. Well, they technically went up two points. Right? GPT five two to GPT five four. Anytime this happens for the most part for that week or two weeks or three weeks, that model's probably gonna be the best in the world, until the next one comes out. I think this is why it's important for all business leaders to well, number one, stick to your AI operating system of choice. You can't be trying to switch your workloads every single week because, essentially, every single week or every other week or at least once a month, there's gonna be a new world's best model. So, let me know. On Wednesdays, we go hands on with our AI at work on Wednesdays.
Jordan Wilson [00:13:04]:
Should we do the g p t five four model? So we did more of a coverage. Here's the news. Here's the seven big takeaways on Friday's show. So if we go hands on on Wednesdays, let me know. Should we do that? Should we do something else? Because maybe this one is just as important. Our next piece of AI news, yeah, might sound small, but it's actually kinda big. So Chad GPT announced Excel. Yeah.
Jordan Wilson [00:13:29]:
Chad GPT for Excel is rolling out in beta, offering a major upgrade for professionals who spend their spend their days in spreadsheets like me. So the new Excel add in for Chad GPT embeds CHAD GPT directly into Excel workbooks, enabling users to build, update, and analyze models using plain language instead of manual formulas. So powered by the new GPT five four model, the tool is designed for advanced financial reasoning in Excel based modeling with performance nearly doubling on internal investment banking benchmarks compared to previous versions. So users can now integrate trusted financial data from providers like FactSet, Dow Jones, LSEG, S and P Global, Moody's, etcetera, directly within Chatt GPT without having to leave. So streamlining streamlining research and analysis. So Chatt GPT for Excel supports tasks such as scenario analysis, reporting, inventory management, and budgeting, preserving workflow structure, and formulas while speeding up workflows. So the tool can explain outputs, trace errors, and link answers to specific cells, making it easier for teams to audit and verify results before making decisions. So who can get it? Well, right now, access is currently available in beta for Chad GPT business enterprise, EDU, teachers pro, and plus users in The US, Canada, and Australia.
Jordan Wilson [00:15:00]:
Alright. And that's not it. Maybe you're a Google Sheets person. For me, I probably use Google Sheets a little bit more than Excel, although I do use Excel, now and then. But, OpenAI did also announce that they're gonna have a similar add in coming for Google Sheets soon. So, teams can also export cited outputs like earnings, summaries, and valuation snapshots to PDF or Microsoft Word. This one, it's pretty interesting, because although Microsoft, did finally kind of roll out, you know, Copilot in Excel that actually worked a couple of months ago, I think it was maybe Claude's plug in that really set this off. Right.
Jordan Wilson [00:15:43]:
I think that was in February is when it started to roll out more broadly. So here we are next month. It seems like I don't know. It seems like OpenAI has been kind of tracing in some of Anthropic's footsteps. You know, I talked about that a little bit last week, and I think in our start here series this week, I'm pretty sure we're gonna do kind of the, the AI race. So giving everyone kind of a baseline of where all the different companies are at, and I think that's something I'm gonna be addressing, in that episode. So if you are curious who's winning in what area, right, if you use Microsoft Excel, should you be using Copilot? Should you be using the Claude plug in? Should you be using the chat g b t Excel add in? Right? Those are the types of things we're gonna be addressing on that show. Alright.
Jordan Wilson [00:16:32]:
OpenAI was not the only company with some big releases because Google actually had a small release, but it is big. That's because they launched their Gemini three one flashlight, their newest AI model, which is now the fastest and most cost effective in their lineup offering in the Gemini three series. So the key improvement here is speed. So if you've never used Gemini three one flashlight, it is, well, like, you would think by the name, flash and light. It is a lightweight version, and it is freaking fast. So, Google says that Gemini three one flashlight delivers 2.5 times faster time to first token than its predecessor Gemini 2.5 flash in, outputs 363 tokens per second up from 249. So not only is it more responsive, well, it is just straight up faster. So this model is designed for instant real time applications like customer support, content moderation, and user interface generation where even a two second display, delay could disrupt the user experience.
Jordan Wilson [00:17:49]:
Also, Flashlight introduced adjustable thinking levels, allowing developers to balance speed and reasoning depth for different tasks from rapid classification to complex code generation. So right now, benchmarks, pretty good. Right? So Flashlight, did achieve an ELO score of fourteen thirty two, on ARENA with standout results in scientific knowledge, multimodal tasks, and multimodal q and a. It's also, well, much cheaper. So for enterprise use, flashlight is priced at only 25¢ per million input tokens and a dollar 50 per million output tokens, making it significantly cheaper than the smaller competitors like Claude's, Haiku, and even their previous Gemini models. So flashlight is kind of positioned as the reflex of the Gemini lineup, handling the high volume repetitive task at scale, while its sibling Gemini three one pro focuses on deep reasoning and complex problem solving. So this is one of those things I think as more and more people become software developers, right, which I know that might sound crazy, but even the leaders in software development right now are saying they don't write code. Right? They talk to agents who write code.
Jordan Wilson [00:19:08]:
So I think that really levels the playing field of who can be a developer now, who can build software in the future, and I think we also have to think and wonder about, well, what does software even mean for the future? But regardless, I think this model, it has to be on your radar. Right? The example there, customer service. Right? You probably shouldn't be using Gemini three one Pro or GPT five four thinking or, you know, Opus 4.6 for things like customer service. So maybe if you in implemented something at your company, you know, six months ago, a year and a half ago, this might be the model to look at when it's like, is it worth updating? And, well, I think that Gemini 3.1 flash light has to at least, be on the radar at only one eighth the cost of pro. So for certain tasks, I think it is a must have. Alright. Next piece of AI news, did not see this one coming. So Open Claw is making headlines from its newest fan, and that is NVIDIA CEO Jensen Wong.
Jordan Wilson [00:20:18]:
As he declared Open Claw, quote, unquote, ready, probably the single most important release of software probably ever, end quote, at the Morgan Stanley Technology Media and Telecom Conference. So pretty big praise from the CEO of the most valuable company in the world, an open source project that hardly no one had heard of, three months ago, four months ago. Now he said is probably the single most important release of software ever. So, OpenClaw achieved in just three weeks the adoption, that took Linux thirty years, making it the fastest growing platform ever for any open source project. So the platform's GitHub stars and download counts have surged recently, outpacing Linux's long term trajectory and setting new records for community engagement. So OpenClause agents differ from traditional chatbots by focus on by focusing on autonomous actions rather than simple queries allowing them to perform complex tasks such as researching, writing, and running workflows. One highlighted that these agents, right, it's like, okay, why is, the GPU king talking about autonomous, you know, open claw? Well, that's because the agents consume 1,000 to 1,000,000 times more tokens than standard chat interfaces. Yeah.
Jordan Wilson [00:21:52]:
Driving unprecedented demand for compute infrastructure. So yeah. If you're wondering why is the CEO of NVIDIA talking about OpenClaw, well, it's because OpenClaw gobbles up a lot of tokens and, well, what needs or what do, tokens need or what do all these models that eat up the tokens need? Well, they need NVIDIA's GPUs. So there's the connection. So if you don't know Open Claw, well, what rock have you been hiding under? But the project's brief history so far includes operating under names like Claude bot first until they got a nice legal letter from Anthropic, probably not a good move, Anthropic. Then they were called Moltbot temporarily, and now they're called OpenClaw. And, of course, a couple of weeks ago, OpenAI kind of acqui hired the project. Right? They, hired the sole, developer of OpenClaw.
Jordan Wilson [00:22:46]:
So now it's kind of under the OpenAI umbrella. But Wong's comments signal a shift in AI development from information retrieval to autonomous task execution, which could reshape how businesses and individuals are using AI. And I think it's already reshaping that. Right? Because we've seen even in the last few weeks, interestingly enough. Right? I think perplexity's computer, is a version of OpenClaw. I think, anthropic, we're gonna talk about some of their, releases here in the what's new and what's next. I think they had two updates that are kind of, you know, trying to help them maybe better compete against Open Claw. So, yeah, it's been interesting so far, to see not only the crazy user adoption, but also the big companies are making shifts to make their products more autonomous and easier to use, kind of like OpenClaw.
Jordan Wilson [00:23:45]:
Alright. Our next piece of AI news, OpenAI is reportedly developing its own code repository platform to address frequent GitHub issues according to the information. So OpenAI engineers have faced multiple GitHub outages, sometimes lasting several hours, disrupting their ability to commit or collaborate on code. So because of that, apparently, OpenAI might be building their own version of GitHub, which is crazy because the entire coding industry runs on GitHub. And I'd say there's really no wide stream alternative. Well, maybe that could change. So the project is still in its early stages according to reports and is expected to take several months before completion. So OpenAI is considering whether to offer the platform to its customers or just keep it exclusive for internal use.
Jordan Wilson [00:24:42]:
So the move is a response to reliability concerns as GitHub is owned by Microsoft, a major OpenAI partner, but we've seen Microsoft also over the last couple of months since OpenAI's kind of transition from a, nonprofit to a public benefits corporation or PBC. Right? We've seen Microsoft start to incorporate more, Claude models, Microsoft investing in anthropic. So it might seem like OpenAI is saying, okay. Well, we're gonna start you know, they've obviously part partnered, with some other big players that are Microsoft competitors, and they might just be, well, starting their own projects that might compete with Microsoft directly. So the new platform, at least right now, aims to ensure OpenAI's engineering teams have continuous access to their code, reducing downtime and dependencies on external services. If OpenAI does decide to sell access to others, it could provide a new alternative for companies seeking more reliable code hosting and collaboration tools because here's here's the reality. As more and more people, I think with the, the recent popularization and explosion. Right? I think maybe Cloud Code in December, January started it, and then we have OpenAI's extremely popular codex.
Jordan Wilson [00:26:02]:
Right? I do think GitHub's well, I can only, assume their servers are legit melting with more and more people committing more and more often. And then when you talk about autonomous, right, autonomous coding, it's like, you know, my agents are constantly pushing things to GitHub even when I'm not in front of them telling them to do it. So I kind of understand both sides from this. I can see how Microsoft is probably, struggling, to scale to meet the demand because it's unprecedented. And I can also see why OpenAI might be wanting to offer a competing service. Alright. In our last big piece of AI news this week. Yeah, all the jobs AI might take.
Jordan Wilson [00:26:48]:
So Anthropic released a new study that warns AI is on track to disrupt a wide range of white collar jobs. So according to the study, AI models like Claude are already capable of performing up to 94% of tasks in computer and math jobs, but are currently being used only for about a third of those tasks in real world settings. So the gap between what AI can do and what it is actually doing is described as vast, with researchers predicting that as adoption increases, AI will take on more professional tasks, especially in business, finance, management, legal, and office administration roles. So the workers most at risk are not those in manual labor, but rather highly educated, well paid professionals, such as lawyers, financial analysts, and software developers. So the study found that those in the most AI exposed jobs are 16 percentage points more likely to be female, earn 47% more on average, and are nearly four times as likely to hold a graduate degree. So the reports introduces the concept of observed exposure, comparing AI's technical capabilities to its actual use in the workplace and finds that AI is barely being tapped into for what it can actually do. So for jobs requiring a physical presence such as cooks, mechanics, bartenders, AI exposure remains near zero according to the report with little risk of automation in the near future. The study warns of the possibility of a great recession for white collar workers if AI adoption accelerates drawing parallels to the job losses seen during the twenty o seven to twenty o nine financial crisis.
Jordan Wilson [00:28:43]:
Recent US labor data shows a slowdown in hiring for AI exposed fields such as the ones, mentioned in Anthropic's study with a 14% drop in job finding rates for young workers since the rise of tools like chat g p t, though there has not been a systematic increase in unemployment yet. So I have a lot of thoughts on this study, and it's something that I've been talking about for a very long time. Right? I even dubbed it, this gap that they're talking about. I talked about it, on our 2026 AI prediction and roadmap series, and I also technically predicted the exposure to these, kind of high income professional services, two years ago. So I'm actually gonna be tackling this, on tomorrow's show because I think it's worth a deeper dive on this because there's a lot of findings that I don't really have time to get into in a short two or three minute recap. So, for more on that show, make sure to tune in tomorrow. Alright. So that's it for our big stories, but each week we bring you what's new and what's next because there's obviously a lot more going on.
Jordan Wilson [00:30:00]:
Right? We've been doing, well, we've been doing the Everyday AI Show for three years, but we've been doing our, AI news show for almost two years. And two years ago, you know, there's maybe 10 stories that were consequential. Now there's, like, 30 to 40 each week. So we usually focus on eight to 10 big stories and then our what's new and what's next. These are, well, maybe smaller new stories, new AI features, rumors, leaks, all of that good stuff. So let's get straight into it because there's a lot. Alright. Ready.
Jordan Wilson [00:30:31]:
So Google shipped a new command line interface tool for Google Workspace, which sounds technical, but it's absolutely insane. Anthropic launched clawed marketplace, letting enterprise spend commitments on partner built clawed tools, so that's modeled after AWS. Speaking of AWS, they put a private self hosted Open Claw agents on Lightsail. Microsoft's SharePoint admin agent is now fully available. Infraffic released scheduled task for clawed code. Yeah. Remember I said they're kinda going full open claw mode? There you go. They released scheduled tasks and also a loop feature, over the weekend.
Jordan Wilson [00:31:10]:
So, yeah, bringing the open claw features to clawed code. According to reports, OpenAI topped $25,000,000,000 in annualized revenue. Google released Canvas in AI mode. Y'all go check it out. If you haven't ever used Google Gemini Canvas, it's now available in AI mode. So, yeah, go use it. I love it. OpenAI launched codex security and research preview, which was formerly called Aardvark.
Jordan Wilson [00:31:38]:
Meta signed huge chip deals with NVIDIA and AMD, boosting their AI spending. OpenAI revised their Pentagon contract, banning surveillance and autonomous weapon use amid backlash from employees and just the general community. OpenAI released codex for Windows. Although I don't have a Windows machine, I actually got one. Never set it up. But, a lot of people have been saying the Windows version is not quite as stable or useful as the Mac version yet, so we'll see how that pans out. Chad GPT, here's a big one. Might sound as small but huge.
Jordan Wilson [00:32:15]:
So Chad GPT finally released skills in Chad GPT, but only for users on business enterprise, EDU, teachers, and health care, and the codex in the API. So essentially, everyone now has access to skills and chat g p t except free users and users on the $20 a month plus plan. Yeah. And the pro plan. I didn't even really realize this, until right before, this show when I logged out of my pro plan and logged into my, business plan. And I'm like, oh, skills are here. Sweet. Alright.
Jordan Wilson [00:32:47]:
Speaking of OpenAI, they updated their prompting guide for GPT five four. We'll put that in today's newsletter, so make sure you go check it out. Quinn released their Quinn 3.5 medium model. So, the, popular, model maker at Alibaba, but also a bunch of their leadership left. So we'll see what actually happens with Quinn in the short term after that. Perplexity rolled out skill support for their new computer products. I've been loving computer. Did a, sounds like Ron Burgundy.
Jordan Wilson [00:33:21]:
Like, I love lamp. I love computer. The problem is the credits eats them up. Yeah. We covered that, last week if you wanna go check that show. Amazon just launched Connect Health, letting businesses spot and fix contact center issues in real time. Microsoft announced support for GPT five four inside of Copilot, so no big delay there for a Microsoft. It is available, rolling out now.
Jordan Wilson [00:33:44]:
Speaking of Microsoft, they're also testing a version essentially of Chad Chibiti's Pulse, which is a personalized news feed inside of Copilot called Copilot Discover based on your history. The US Supreme Court declined hearing whether AI art, generates sorry. AI generated arts qualifies for copyright. OpenAI rolled out support for saved prompts inside of its Atlas browser. NVIDIA announced that they're no longer gonna be investing in OpenAI and Infropic after the companies go public, which is expected to happen later this year. Google released cinematic videos in notebook l m, and my gosh, they are so good. Only for ultra subscribers right now, so you gotta be on that $200 a month plan. OpenAI is reportedly working on a bidirectional audio model for better human to AI voice interactions.
Jordan Wilson [00:34:36]:
I really hope OpenAI, updates their voice mode in chat gbt. I hope they update their agent mode, at least to gbt five four. A lot of great, multimodal tech that OpenAI has that seems like it hasn't been updated in a while. In a White House meeting, tech giants pledged to cover grid upgrade costs for AI dentists, AI data centers in The US. So essentially saying they're not gonna pass the electricity bill, to local residents where they're building, these new AI data centers. And then last but definitely not least, small footnote, big footprint. Microsoft released five for reasoning, their 15,000,000,000 parameter open multimodal model, balancing reasoning, accuracy, and efficiency. Alright.
Jordan Wilson [00:35:22]:
So that's it. That's our big AI news stories. That's our what's new and what's next. But we did start something. Alright. Last week, kinda threw this out there. Seemed like most of you liked it. So here's the problem with the show and how it's set up here at Everyday AI.
Jordan Wilson [00:35:39]:
On Mondays, we do exactly this. Right? We talk about the big news stories, and then at the end, we go over all of these bullet points like I just did. What I've learned to realize is a lot of those bullet points, you all probably wanna know more about. Right? So, yes, on Wednesdays, we always go hand hands on, but with one new AI feature, one new AI update. So we threw out this concept of doing a, more hands on, kind of update on actual AI features. Right? So as an example, in those, what's new and what's next, probably a lot of those smaller things are going to be the things that actually move the needle for your business. And I can't address them all in, you know, the AI news and on our one kind of hands on show on Wednesday. So I started to think, okay.
Jordan Wilson [00:36:32]:
Do Do you guys want this, you know, going over a handful, maybe five to eight of these new features inside of Chad GPT, Copilot, Claude, Gemini, maybe perplexity, maybe one or two others. Is it something you want? If so, let me know. I don't know. Say, I always should have, like, a word to tell you guys to comment, and I don't even know what I'm gonna call it. How about this? Give me a name for it. If you want something, maybe it's called Friday features, and we put it out on Fridays. I don't know. It's up to you.
Jordan Wilson [00:37:03]:
I work for you. If that's gonna be helpful, leave me a comment. If you're listening on Spotify, leave a comment there. If you're listening on the livestream, leave a comment. So, yeah, we did that show, last Thursday. So if you do wanna go, check that one out because as an example, some of those bullet points that we talked about there at the end, the, you know, the cinema overviews inside of notebook LM, you you know, the, workspace CLI, which I think is huge. So we did that on Thursday, episode seven twenty seven, so make sure you go check that out. Alright.
Jordan Wilson [00:37:33]:
That's a wrap. A lot happening this week. I hope that our Monday shows are helpful to put you in the driver's seat. Feel confident for the week. If you're making decisions on front end AI strategy and implementation, well, you gotta know what's going on. So if this was helpful, make sure you tell someone about it, and then make sure you go to your everydayai.com. Sign up for the free daily newsletter. Thanks for tuning in.
Jordan Wilson [00:37:55]:
We'll see you back tomorrow and every day for more everyday AI. Thanks, y'all.