Everyday AI Podcast – An AI and ChatGPT Podcast - EP: 478 OpenAI's $20,000/mo agent, Manus impresses, Google's new AI Mode and more AI News that Matters

Episode Date: March 10, 2025

An AI agent that costs $20,000 a month? 🤑Are Microsoft and OpenAI fighting? 🥊And why does Meta have its eyes sets on hundreds of millions small businesses? 😧Don't waste hours each day trying to figure out what the latest in AI developments mean for you or your company. Instead, join us on Mondays for the AI News That Matters. 👇Newsletter: Sign up for our free daily newsletterMore on this Episode: Episode PageJoin the discussion: Thoughts on these stories? Join the conversation Upcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: [e-mail] with Jordan on LinkedInTopics Covered in This Episode:Microsoft and AI Sales AgentsReflection AI and FundingManus by Chinese Startup MonicaMeta's AI VisionAnthropic's Proposal for AI and National SecurityGoogle's AI Mode in SearchAlibaba's QWQ 32B ModelOpenAI's ChatGPT Desktop App UpdateOpenAI's $20,000 Monthly AI Agent PlanMicrosoft's Exploration of AI Options Beyond OpenAITimestamps:00:00 "AI Innovations and Industry Shifts"03:47 Salesforce vs. Microsoft: AI Agent Battle06:24 Reflection AI's Autonomous Superintelligence Vision10:55 Manus AI Challenges OpenAI13:23 Early Access Tech Review Prospect16:24 Meta's Expanding AI Vision22:54 Google's AI Mode: Enhancing Search27:18 Small Model Beats Giants in AI28:42 Chinese AI Models Surpassing Western Rivals34:44 OpenAI Launches Premium AI Agents36:21 OpenAI's Potential Revenue Shift39:27 Advanced GPT-5 Model Insights43:05 Microsoft's Backup AI Plan48:27 AI Advancements Span China to Silicon Valley50:51 Subscribe for Daily AI UpdatesKeywords:autonomous agent, Meta AI, OpenAI, Microsoft Copilot, AI innovation, AI news, NVIDIA GTC, Reflection AI, Salesforce CRM, AI sales agents, Google DeepMind, AI coding tool, artificial general intelligence, China's AI agent, Claude model, Alibaba QWQ model, AI business applications, classified info sharing, Google AI mode, Anthropic, Meta AI agents, open source AI, ChatGPT desktop app, professional AI agent, AI security, AI development, Microsoft AI strategy, Llama, generative AI, AI startupSend Everyday AI and Jordan a text message. (We can't reply back unless you leave contact info) Start Here ▶️Not sure where to start when it comes to AI? Start with our Start Here Series. You can listen to the first drop -- Episode 691 -- or get free access to our Inner Cricle community and all episodes: StartHereSeries.com Also, here's a link to the entire series on a Spotify playlist. 

This is the Everyday AI Show, the Everyday Podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business, and everyday life. Meet Firefly AI Assistant, now live in Adobe Firefly, the all-in-one creative AI studio. Just describe what you want to create and the assistant handles the rest, orchestrating multi-step workflows across Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome. The assistant accelerates execution. There's a new super smart autonomous agent that might be better than anything any other AI company has to offer.

Meta is going after hundreds of millions of businesses with their next AI innovation. Open AI would you pay $20,000 a month for a super smart future model? And are Microsoft and Open AI kind of breaking up? Or is it just Microsoft diversifying options for its co-pilot platform? All right, we're going to be answering those questions and hopefully a lot more on today's edition of Everyday AI. What's going on, y'all? My name is Jordan Wilson, and I'm the host of Everyday AI. This is your daily live stream, podcast, and free daily news that are helping everyday people not just learn AI, but how we can all actually leverage.

it to be the smartest person in your department or in your company when it comes to AI. So if you want to not just grow your company, but grow your career with AI, you are in the right place. If it's your first time, thank you for listening or watching. So yeah, if you're on the podcast, please make sure to subscribe, click that subscribe button. If this is helpful, leave a rating or maybe come join us live because this is a live stream. So what's up and good morning to our live stream audience, Michael and Sandra on YouTube, Brian, Samuel, Rolando, Joe, everyone else. Thank you for tuning in. So almost every single Monday, we bring you the AI news that matters. So you shouldn't do this. You shouldn't spend hours every single day

keeping up with AI. You should actually just be using the best of it and leave us to the rest of it. That was unscripted, by the way. But, you know, we do this every single day. And on Monday, we essentially say, yo, here's what's happened this week. Here's what's coming up. Let's cut through the fluff and just tell you what actually matters. So, you know, yeah, if you can't maybe join us every single day or spend hours on your own exploring what the AI world has to offer this on every single Monday, that's when you should definitely be joining us.

Also, if you haven't already, please go to your everyday AI.com. That's where you can sign up for our free daily newsletter. Because, yeah, aside from the live stream of the podcast, we recap each. podcast in the newsletter as well as bringing you insights from everywhere else across the web. All right. And as a reminder, I will be out next week in San Jose at the NVIDIA GTC conference. So partnering up with NVIDIA, I'm excited about that. So if you're going to be at the GTC conference in NVIDIA, let me know.

InVIDIA essentially powers the whole generative AI movement. So there should be some pretty exciting happenings happening there. and make sure you tune in because we're going to have some special podcast interviews with some leaders at NVIDIA and others that you're not going to hear anywhere else. All right. So make sure you tune in next week. All right. Let's get into the AI news that matters for the week of March 10th. Let's get it. All right. So first, Microsoft unveiled some new AI sales agents to challenge Salesforce.

So Microsoft introduced a pair of AI-driven. sales agents aimed at automating sales tasks and directly competing with Salesforce CRM AI offering. So yeah, Salesforce has kind of been picking a fight with Microsoft, I think intentionally, to try to grab headlines and to try to carve out their space in the autonomous agent space. So yeah, Salesforce has their agent force, autonomous AI agents. And, you know, Microsoft is now clapped back. So their new sales agent in sales chat are AI tools that can identify leads, schedule meetings, follow up with customers, and provide sales insights by analyzing emails, CRM records, and other data.

So the agents integrate with both Microsoft Dynamics 365 and also Salesforce's platform, letting sales teams work deals even without opening their CRM software. So Microsoft says they can be fine-tuned on company-specific data for more accurate tailored responses. So the new AI sales agents will enter public preview in May, this May, so about two months here, via Microsoft 365 copilot, expanding the copilot ecosystem from mere assisted prompts to fully autonomous task completion in sales workflows. So to encourage adoption, Microsoft also launched an AI accelerator for sales program to help businesses build and implement those agents, part of an effort to lure customers away from quote,

and quote legacy CRM vendors like Salesforce. So, you know, I'm interested to see how this is all going to pan out. Also, live stream audience, let me know what you want to hear more of. I'm probably going to bring on some Microsoft leaders in the coming weeks because I think they've really been crushing it, right? So not just what we have here from Salesforce, but, you know, they have their new co-pilot dragon in the health space. Their co-pilot free platform has gone.

bonkers just like they're now offering for free what even other companies are making you pay for. So pretty exciting announcements coming out of Microsoft. All right. Next, a new venture called Reflection AI has emerged from Stealth with $130 million in early stage funding aiming to develop super intelligent AI systems. So this is from some former researchers. at Google DeepMine and some other places, but they've raised $130 million across two different rounds, a $25 million seed led by Sequoia Capital and CRV, and $105 million series A,

co-led by CRV and light speed venture partners. So the company's backers include tech heavyweights like NVIDIA's venture arm, LinkedIn co-founder Reid Hoffman, and Scale AI CEO Alexander Wang. All-Star roster, sorry, an all-star roster reflecting high confidence in reflection AI's vision. So it's founded by X Google DeepMind researchers who both worked on Google's Gemini AI project, which, so as a first step, the team is developing an autonomous coding tool that can handle complex programming tasks without human help. So the founders argue that mastering autonomous coding first will require breakthroughs,

and reasoning and self-improvement that naturally extend to broader computer-based work. So in other words, a coding agent that can teach itself and reason through problems could be repurposed later as a more general super intelligent assistant. So, yeah, end goal is super intelligence. That's always weird now saying out loud when companies start with the end goal to just, you know, get super intelligence when, you know, we technically, depending on what definition you look at, haven't achieved, you know, AGI or artificial general intelligence, which is essentially when a single AI system is, you know, smarter than most all humans across all tasks. So pretty, pretty interesting now

to see not just reflection AI raising some big money, but also SSI Safe Super Intelligence Incorporated from Ilius Satskiver. So pretty, pretty interesting. A lot of money and big AI brains. you know, working straight in this super intelligent space, which I guess is better, right? I guess it's better that we have super smart people working on, you know, hopefully safe super intelligence and not just, you know, waiting until we quote unquote to cross the finish line for AGI. So I guess it's nice that, you know, people are looking at what comes after AGI because if you listen to the show, I did a show about a year ago.

Maybe I'll have to do it again. But this whole artificial general intelligence debate, I don't know. It's kind of strange because the goalposts are just going to continually be moving as AI systems become more and more capable. You know, if you look at definitions from 15 years ago, right, this is what I did. I spent way too many hours going to look at definitions from, you know, the year 2005, 2008, 2010, right? If you look at those older definitions from 15 plus years ago, we've definitely achieved

artificial general intelligence, but it's a working definition. It's fluid. It always changes. But, you know, I guess there's plenty of money to be working in super intelligence. All right. I'd say one of the more exciting stories of the week is probably China's fully autonomous AI agent Manus. So it debuted and it is powered, FYI, and this was confirmed by Manus's and I don't know if it's Manus or Manus, I guess we'll find out. I haven't had time to play with it yet because it is currently invite only, but it is powered by Claude and also Alibaba's QWQ model. So a Chinese startup named Monica has unveiled Manus,

calling it the world's first autonomous AI agent capable of executing complex tasks without human intervention. So the debut sparked buzz as Manus reportedly outperformed other AI assistants on key benchmarks, at a new leap in agent capabilities. So in demos, Manus handled real world tasks from end to end, from screening job resumes to conducting property research, all on its own dedicated computing instance. So it can browse the web, write and debug code, generate images, and even perform gigs on freelance platforms like Upwork and Fiber

without human guidance. So the team behind Manus claims it can achieve state-of-the-art results on different benchmarks surpassing general AI agents like ChatGPT and Google Gemini in similar evaluations. So like I said, Manus is currently invite only, but the developers plan to open source its underlying models later in 2025 to foster transparency and common collaboration. So under the hood, Manis leverages a combination of cutting edge AI models. So reports and, well, their own team has confirmed that it uses Anthropics Claude, but

Claude 3.5, interestingly enough, I'm guessing the Claude 37. The Manas team was probably already too far in the development. So we'll see if it does upgrade to Claude 37 or maybe uses a different model. But it does use Claude 3.5 for as well as Alibaba's QWQ. to carry out its tasks. So the emergence of Manus highlights the rapid progress of independent AI labs. And this is pretty, pretty, I think we're going to be talking about this a lot in the next couple of weeks, mainly because it almost seemingly competes with multiple of open AI's highest tiered kind of products or models. So it does a lot of the kind of deep research, which I

I think is open AI's deep research, I would say at least for the average person, right? The average person is not going out there and, you know, coding for many hours a day, right? So, you know, for different fields or different job types, different verticals, I think there's maybe more impressive AI tools or AI systems out there. But for the average everyday person, I think across the board, Open AI's deep research is by far the most capable and the most powerful AI tool anyone can use right now. But this manis could change that. Well, we'll see once it's available to everyone because it does have a little bit of, you know,

the deep research vibes. It can go off and do deep researching tasks on its own. But also it does kind of have operator vibes as well. So, you know, chat GPT operator is an autonomous agent that can go out and perform tasks in a virtual environment. So it's almost like Manus does kind of both, right? And like I said, it has kind of this two model approach under the hood. You know, it has the kind of transformer model of Claude 3.5.

And then the QWQ 32B from Alibaba to do some reasoning. So it looks fairly impressive. Early reports, like I said, it's invite only. I don't have access yet. But if I do, if you all want to see a show on this. if and when it either is open to the public or if I do get access, let me know and I'll give it the usual deep dive. But it could be a real contender here, right?

Especially, I think, as the team explores early feedback, right? So, yeah, it just kind of launched, I think a couple of days ago, very limited invite only. But like I said, it seems like it kind of combines I think there's three main pieces to dissect here. One is it does use this hybrid model approach using both Claude and QWQ. Two, it does have this kind of deep research capabilities to go off and do a lot of research for people. But then also, it is more agentic like operator, right?

It can go out and actually just perform tasks for you autonomously without human intervention on your behalf. So it can go out and code things. They can go out and create documents, right? It can go out and perform actions for you. So it's kind of like a combination of three different key pieces that something like ChatGPT does right now. Yeah, Michael says, somebody invite Jordan WTF Manus. Yeah, I agree.

Someone dropped me an invite. Adobe just introduced an entirely new way to create, bringing the power and precision of its creative suite into one conversational experience. Meet Firefly AI Assistant, now live in the Adobe Firefly app, the all-in-one creative AI studio. Powered by Adobe's creative agent, Firefly AI Assistant lets you start with your vision, just describe what you want, and shape the outcome as it takes form with the Assistant. The Assistant orchestrates multi-step workflows, drawing on 60-plus pro-grade tools across Adobe Creative Cloud apps, including Photoshop, Illustrator, Premier, Lightroom Express, and more to help bring your ideas

to life. You can also get started with creative skills, a growing library of pre-built workflows for common creative tasks, like batch editing photos, creating mood boards, portrait retouching, and creating social variations. Every step the assistant takes is visible so you can refine, redirect, or take over at any time. You stay in the driver's seat as the creative director. Adobe Firefly AI assistant now in public beta. See it today at firefly.adobie.com. Let's keep going. Speaking of invites, Meta's like trying to invite hundreds of millions of businesses

to use Lama. So Meta is predicting that hundreds of millions of businesses will adopt to their AI agents. So Meta's head of business AI, Claire, she has said the company is gearing up for a world where virtually

every business uses AI agents to interact with customers. So in an interview, she noted meta's AI tech already reaches 700 million individuals, you know, through their other arms, right? So through Facebook, WhatsApp, Lama, right? Like META has a little of everything, Instagram, right? And they drop, you know, Lama and their AI smart assistance everywhere. But it sounds like their goals are much more than that because meta is focusing, especially on small businesses, yeah, hundreds of millions of small businesses that lack in-house

AI teams. So Meta's idea is to provide affordable AI agents through their current platforms like WhatsApp, Facebook, and Instagram that can answer customer questions, remember preferences in automate repetitive tasks 24-7. So this could democratize AI access to much smaller, you know, small businesses, right, that might just be side hustles that are operating on WhatsApp, something like that. So pretty interesting approach here from META. So the company's confidence is reflected in its investments. So META has been expanding its AI offerings, positioning itself as a key provider of business, AI for marketing, customer service, and beyond. So if META's vision holds true, AI agents could soon handle frontline customer interactions for millions of merchants globally.

So yeah, then all us humans can, you know, I guess go scroll, you know, Facebook and Instagram and WhatsApp, you know, while the autonomous AI agents, you know, slide into everyone's DMs and, you know, push, push your merch or something like that. So, yeah, keep an eye out on meta. They have their first ever Lama Khan coming. up pretty soon, which I'm excited to see that is in April. So we might be seeing Lama 4. You know, some rumors are saying we might see something in the autonomous AI agent's space. So pretty big announcement coming up from Mata as we kind of gear up into the spring announcement season for AI companies. So yeah, starting with NVIDIA GTC, then we'll have META's LamaCon as well as a few other big kind of tech and AI conferences coming up.

All right. Let's keep this AI train rolling to chew. So pretty interesting that I didn't see a lot of people talking about this, not even a lot of news coverage, which I found interesting, but Anthropic is urging the creation of classified info sharing for AI companies with the U.S. government. So in a policy proposal, AI startup Anthropic is calling for new classified communication channels between AI companies and the U.S. government.

So the aim is to securely share information about advanced AI developments, reflecting concerns that powerful frontier AI systems could pose national security risks. So Anthropics recommendations came in a 10-page response to the White House's request for public input on AI, on their national action plan. So the company warned that AI capable of matching or exceeding Nobel-level human intellect could arrive as soon as next year, yeah, 2026, and argued that officials need better tools to monitor such rapid progress. So among the ideas, Anthropic proposes establishing classified channels so that top AI labs

can discreetly share sensitive details of their most advanced AI models with government security agencies. So this would be complemented by expedited security clearances for key industry scientists ensuring the government isn't caught off guard by breakthroughs known to corporate insiders. So Anthropic, led by CEO Dario Amati, liken the situation to arms-controlled coordination, suggesting that secrecy and tight coordination may be needed. to manage AI capable of unprecedented reasoning or autonomous actions. So Anthropics stance highlights attention. While these classified channels could help U.S. authorities respond to AI threats,

they might also limit public transparency around critical AI decisions. So yeah, I don't know. I don't necessarily see this happening. The current Trump administration has taken a pretty hands-off approach when it comes to AI. They kind of gutted the USAI Safety Institute. So I don't necessarily see this happening, although it probably should, right? When you think about when these companies are developing AGI, essentially now you have labs working on ASI, right? ASI is when it kind of gets scary, y'all.

That's when AI systems develop their own AI that, you know, they don't need humans for, right? that's when it starts to be like, yeah, maybe we should, you know, top AI labs instead of just, you know, fiercely competing with each other. Maybe they should be sharing what breakthroughs they have achieved at least if it could impact national security. If it could impact, you know, the world economy, which, you know, I'm not being a crazy weird AI guy when I say that. That's actual concerns that we should all be focusing on, right? because think of what you, think of what you at peak, you know, human performance can do with an AI system, right? It's a multiplier. It is a compound multiplier, right?

In terms of what you, an individual human can achieve, right? When you get, you know, chat GPTs operator or deep research or manis, when it can do all these things, right? Think of what you, one human can do when you have these powerful AI. systems. Now think these systems that we're playing with and using with now are old news, right? All the AI labs have the next version of these and probably the next version, which are probably scary good. So you have to think of, okay, what happens when and if this gets into the wrong hands? Or what happens to jobs? What happens to the economy, right? What happens to national security, right?

If you have an AI system that could essentially go rogue, right? And it's smarter than literally every single human researcher, you know, in a couple of years. When it's smarter than every single human researcher and it can just build its own version of, you know, can build its own version of operator, right? Like that's something I saw with Manus. Someone used Manus to build a, essentially a fine-tuned version of Manus. Right. So yes, the systems that we have today are capable of building their own systems with human intervention, but what happens when they're way smarter? So I like what Anthropic is pushing. I don't personally think

that this is going to hit, at least with this current administration's more hands-off approach to AI. All right, Google has unveiled AI mode in search, a major shift on how we interact with the web. So Google has introduced AI mode an experimental paid feature that uses advanced AI capabilities to develop more comprehensive answers to search queries. So Google has announced AI mode as a part of its Google One AI premium service, marking the second phase of its AI-driven search evolution. So this feature builds on its Gemini 2.0 model, offering advanced reasoning, multimodal capabilities and real-time. information. So unlike traditional search, Google's new AI mode allows users to ask follow-up questions and provides AI-generated summaries alongside links to relevant websites. However, concerns remain about whether users will actually click the links, potentially reducing traffic

to external sites. And yeah, then we just have this whole, you know, the internet just becomes, becomes AI-generated slop because high-quality content producers go on to business because no one's clicking their links and signing up for their newsletters and using their products and signing up for their services, right? So yeah, there's a lot of downsides, you know, when we have these very capable, you know, tools that can just browse the web and give us literally the pinpoint insights that we need. You know, I'm curious if anyone on the live stream audience has used AI mode from Google yet. I'm thinking we might do a dedicated show on this this week, just because it is a little different than Google's deep research.

I think it's also a little different than Google's, or sorry, than perplexity and GROC and Open AIs deep research. So it is kind of, I think, its own kind of way to interact with the web. So, you know, live stream audience, let me know if you want to, you know, see a breakdown of the new AI mode from Google. So it is designed to query a broader range of websites and simultaneously search related topics, delivering detailed responses. It also includes mechanisms to acknowledge uncertainty or avoid generating summaries when confidence is low. Bless up.

We need more of just AI refusing to answer things if it doesn't know or if it's not super confident. So while Google has promised more prominent links to websites in AI mode, the current version offers fewer links than the demo shown during the announcement. So Google's approach contrasts with competitors like ChatGPT and Claude, which have gained traction for their conversational AI capabilities. So AI modes integration into Google's search index adds real time and local business data as well, differentiating it from standalone chatbots. You know, Marie says, sure.

Sandra says love the demos. Michael says yes. So yeah, we'll see. Maybe we'll do a show on that this week. Maybe we'll ask the newsletter audience as well. I think I intentionally left some shows open because, you know, knowing I'm going to be at Nvidia GTC, I might not be able to cover as many timely things.

So this week kept a couple slots open so we could hopefully, you know, cover some of these newer announcements. All right. Speaking, we already mentioned QWQ 32B. What a name. But Alibaba's QWQ32B model has topped some open source AI charts. So QWQ 32B, I'm going to stop saying that. I'm just going to say QWQ, I guess, which isn't any better.

We really just need shorter names for all these models. But it quickly shot to the top of some open source AI. rankings showcasing a breakthrough inefficiency and sparking optimism for small but mighty AI models. So QWQ32B is short for Quinn with Questions, which is it actually short with that when it's way more syllables. I don't know. But it is a new large reasoning model released under the open source Apache 2.0 license from Alibaba. So it's allowing anyone to use or modify it freely.

So despite its relatively compact size, which the 32B means 32 billion parameters, QWQ demonstrated performance on par with much bigger reasoning models, such as DeepSeaks R1, which has 671 billion parameters. So yes, let me repeat that. So a model about 5% of its size of Deepseeks are 1. QWQ 32B is from, so. Certain performance and certain benchmarks already on par with a much larger deep seek R1. We do see rumors that we might see a deep seek R2, their latest reasoning model in the coming

week. So we'll keep an eye out for that. So Alibaba achieved this by focusing QWQ on advanced reasoning and self-refinement techniques. So the model can generate intermediate questions and reflect on its answers during inference, which makes it excel in domains like math. problem solving and code generation. In fact, the company says, you know, that QWQ leads or ties open AIs and deep seek models in five major benchmark tests for reasoning.

So within days of launch, QWQ 32B was ranked number one on at least one global open source model leaderboard. The news also gave Alibaba stock a huge boost as investors saw the model. success as evidence that Chinese AI innovation can compete head-to-head with Western rivals on quality and efficiency. What's weird about this, y'all, is before any of this, you know, QWQ32B, before DeepSeek, before all of this, I told y'all in my 2025 AI roadmap that by the middle of the year, that Chinese models would probably overtake most U.S. models, and that more than half of the top 10 models, or at least half of the top 10 models, would be from

at Chinese AI company. So we may see that by April or May, right? We might not have to wait until the middle of the year. But I told you all this was coming. So you shouldn't be shocked. All right. Our next piece of AI news, chat GBT's desktop app, has gained two-way coding integration. So this is another, I think, small news that's actually kind of big.

And it also kind of slipped under the radar. This was just essentially announced on. the OpenAI developers Twitter account, and I didn't even see a whole lot of news article about it, but let me tell you why I think it's important. But first, here's what it is. So OpenAI's official chat GBT desktop app on Mac has received a major update, enabling it to work directly with integrated development environments or IDEs. So in practical terms, that means chat GPT can now both read, but also modify and write code in a user's editor like Xcode or VS code automatically, rather than just suggesting code that the user must copy over.

So the new feature is part of Open AIs work with apps capability on Mac OS that lets paid users. So yeah, you've got to be on chat GPT Plus, pro teams, and I think the enterprise accounts as well. So it allows you to connect the chat chbtbt desktop app with developer tools. So for example, chat chvety can now scan the code in an Xcode project, not just suggesting changes, but also apply those code edits directly in the IDE with an auto-apply mode. So this is pretty big and it's indicative of this recent shift. So first, we saw Claude Code from Anthropic released a couple of weeks ago. essentially this, but it's more of a terminal tool than anything else. But Claude code looks pretty impressive, right?

And what's interesting is you have all these essentially vibe coding tools, right? We call them that like windsurf, cursor, lovable. I mean, there's so many that essentially allow you to talk to their system. And usually these systems, you can choose which model, right? So you could most people use, you know, Claude Sonnet 3-5, because it's great for coding. Some people might use something like 03 Mini from Open AI, but essentially you have all these AI vibe coding models that you just talk to,

and then they can build you an app, right, in one shot. Super impressive, something you can run on your computer. But now we've seen this shift where it's like, wait, all the AI labs are like, wait, we want that. We want to get in that game. Why can't we own that? So, you know, Claude was first with its Claude code, which is open, you know, one is more open and free because it just runs in your terminal.

You download a GitHub repository and then can run it on your computer. Whereas ChatGBTGPT's version, you do have to have a paid account and you run it through the app. Personally, for me, I like the latter approach because I'm not a super technical person. Although Claude code, it is fairly easy to set up, but you got to be able to, you know, at least, you know, run some terminal commands. You got to understand GitHub repo. pose, right? So you have to have a little more tech know-how to use Claude code, whereas ChatGBTGPT's new feature,

very, very useful. So the work with apps has always been available. So, you know, as an example, you know, I use this a lot for text edit. So, you know, chat-GPTs work with apps. It can just see what's open in my text edits. And previously, it could just, you know, it would, anything that you told it to do, it would reply back in the body of, of the chat GPT app. So now it will instead, you know, update these actual apps, which I think, you know, the main difference here, nothing is technically new.

It just saves you a ton of copy and pacing and really just explanation. So instead of saying like, hey, chat, JBT, here's some code that I'm working on in VS code, right? I'm trying to do this. Could you update the code, et cetera, et cetera, right? And you'd have to copy and paste it from, you know, as an example, VS code into chat. And then you'd have to get the output from chat GPT, go back to VS code.

So now it just works by itself, right? ChatGPT can now read and write. So it goes from kind of a one-way street to now a two-way coding partner. So I would say pretty big news that no one is really paying too much attention to. All right, something people are paying a ton of attention to, our next story. reportedly Open AI is planning a $20,000 a month professional AI agent. Yeah, I'm going to repeat that. I did not, that's not some weird AI glitch.

By the way, I was watching wreck the internet last night. Love that movie. There's a glitch thing. Sorry, I just had to say that one of my favorite songs is Slaughter Race. Anyways, it was not a glitch me saying that. actually $20,000 a month professional AI agent plan reportedly. So according to reports, Open AI is preparing to offer premium AI agents for businesses at prices up to $20,000 a month.

So these advanced AI agents would function like ultra-capable employees performing specialized work in areas like analytics, coding, or research, and the steep price tag signals. The value Open AI believes they'll be a. able to deliver to enterprises. So this is according to reporting from the information and other outlets that OpenAI is working on three tiers of agent subscriptions. So one around $2,000 a month that's aimed at business professionals for tasks like marketing or customer analysis, one around $10,000 a month for advanced software development

tasks and a top tier, quote unquote, PhD level research agent at $20,000. a month. So these AI agents are envisioned to operate with a high degree of autonomy on behalf of a client, essentially acting as a highly skilled digital staff. So OpenAI CEO, Sam Altman predicted earlier this year that we'd see AI agents, quote unquote, join the workforce and significantly boost company outputs. So the new offerings appear to be how Open AI will make that happen by literally productizing expert AI workers for hire. and big money is backing the idea. So global juggernaut soft bank has reportedly committed $3 billion to invest in OpenAI's agent venture for 2025.

So OpenAI projects that these agent services could eventually make up a quarter of its revenue, highlighting just how central agent as a service might become to its business model beyond just API access or chatGPD subscriptions. So if launched, that $20,000, a month price point means these AI agents would cost as much as a senior human employees. So OpenAI betting companies will pay that if the agent can handle work equivalent to an entire team of researchers or developers. So early adopters might include Fortune 500 firms in areas like finance, tech, or science that can immediately utilize an AI expert on complex projects.

So yeah. And we'll see how things like. Manus impacts this, you know, $2,000, $10,000, $20,000 a month price tag. So I think even if we have things like Manus take off and become very popular in our free or low cost, right? I don't think that it's going to be free in the long run. But I don't think it's going to cost. Manus will cost $2,000 a month. I don't think so.

But I still think even so. I still think even so. Open AI is going to get a lot of customers with that $2,000, $10,000, $20,000 a month. And I think we have to start thinking of it like this. Don't think of it compared to an AI subscription. Think of it compared to paying an employee, right? Because the smartest, you know, the most in-demand employees right now, you know, which are generally working at AI labs, they cost way more than $20,000 a month, right?

I mean, these people are making NBA-like figures, right? Yeah, you have some outliers, you know, making seven plus multiple seven figures a year to work at, you know, big companies like OpenAI, Google, Anthropic, Microsoft, etc. But I would say your quote unquote average everyday, you know, mid-tier software developer is making $400,000, plus a year, which is much. more than the $20,000 a month. So I get that it's immediate sticker shock and people are like, yeah, right? I would not be surprised. I would not be surprised, number one, if this is true, you know, this is just according to reports, but I wouldn't be surprised if Fortune 500 companies are going to pay.

So here's what I think. And I don't have any, you know, inside information on this one at least, right? But what I think essentially is going to happen is think of the best. of Open AI right now. So, you know, GPD 4.5 is a very, very capable model, right? It is ranked as the top model in the world on ELO scores or in terms of what the most humans in the world prefer blindly. All right. So GPD 4.5 is a very capable model. It doesn't have reasoning capabilities. It is serving as a base model for OpenAI's next reasoning model, which will just be coupled under GPT5. So what I see this as, I see this as a way more capable GPT5.

that has reasoning capabilities, that has way better than 4.5 ability for accuracy, relatability, reliability, plus combining it with things like deep research, combining it with operator, combining it with tasks, right? I do think that task piece is something people are overlooking, right? The ability to just schedule that because right now, even with deep research and an operator, you can't schedule those things, right? they still require a human to kick them off. So think GPT5 plus being able to schedule it, plus deep research, plus operator, all in one seamless UIUX, right?

That's ultimately what I think that and the fact that it's U.S. base and I think that a lot of Fortune 500 companies, rightfully so, are going to be hesitant to jump on board with anything from China. I think it's actually going to be pretty popular if this is true, right? I do think that first, Open AI will roll that out in tears. Just like right now, yeah, we have the $200 a month, chat chbtee pro plan, which I think is well worth it, right? But they didn't start with that. They started with a free version. They started with a 20, then went to $20 a month. Then it went to the $200 a month.

So I do think we'll probably start with the $2,000 a month. And then as it gets more capabilities, probably throughout this year and next, then we might see that next step up to $10,000. But I don't know, y'all. What do you think? Kieran is saying, will Open AI replace all of its. developers with 20k AI agents if that's worth it. It's a great question, right? I'm always wondering the same thing.

I'm like, okay, why are all these companies hiring? Because I think right now, even though the cost of compute is going down, they still need all these developers to develop what's next, right? So in theory, once we have ASI, then we won't need to hire, right? Or in theory, we wouldn't need to hire smart humans anymore to improve upon whatever AI models. But until then, you probably need the humans to build what's next on top of your latest and greatest kind of offering. Love what Marie here says, next AI name will be R2D2. That's a better name than a lot of these, you know, O3 mini high and, you know, QWQ 32B.

Yeah, let's just name them R2D2 and things like that. Yeah. Michael says gatekeeping intelligence, starting to support the people calling. them closed AI. I don't have a, I don't personally have a problem with that. But that's just me. And Open AI did, you know, talk about potentially open sourcing some of their future models. So we'll see if that comes to fruition. All right. Last but not least, spicy news. This is soap opera drama. So according to reports, Microsoft is exploring life beyond open AI with both in-house models and looking at other companies to bring in to also potentially power its co-pilot.

So despite its multi-billion dollar investment with OpenAI and reported 49% equity stake, Microsoft is quietly both developing its own large AI reasoning models and also looking at other partners to partner with to power its own own. co-pilot tech. So recent reports suggest that Microsoft has a plan B for AI, which is training internal models, codenamed MAI, that could power products like its co-pilot assistance without relying solely on OpenAI's GPT engines. So according to reporting from the information, Microsoft's AI research group led by Mustafa

Salomon has completed training a family of advanced language models dubbed MAI that perform nearly as well as Open AI or Clod's, inthropic Claude's models on many benchmarks. These models are significantly larger and more sophisticated than Microsoft's current publicly available, as an example, their 5-4 model. So Microsoft has already started experimenting with swapping out OpenAI's models for other alternatives in some of its products internally, and this is according to reports. For instance, it's reportedly tested co-pilot integrations using models from XAI, such as GROC, META's, Lama, and even the Chinese model Deepseek, essentially test driving different AI engines

to see if they can do the job of Open AIs technology currently does in Azure and Office 365. So the motivation is partly cost because running, you know, big models like, you know, GBT45 or OpenAIs, 03 mini high or 01 Pro, it can get expensive. And Microsoft does bear those costs for services like biz chat and their 365 co-pilot. And it's partly strategic as well. So by having its own top tier models, Microsoft could then better negotiate terms, with open AI or eventually deploy more of its AI services entirely on its own proprietary tech. So here's the thing. They haven't technically broken up.

But it's like they're not, they might not be exclusive for too much longer, at least when it comes to Microsoft 365. So yeah, they might be getting in an open relationship with many other AI models, right? So Microsoft, like I said, they haven't officially cut ties. And this is just, you know, right now it's it's rumors. but according to the information, which is usually pretty spot on with a lot of its reporting. So it hasn't cut ties with Open AI yet, but it's actually a major investor, like I said, in Open AI. So Microsoft has all, you know, all reason to continue to support and build and push hand, you know, hand in hand forward with Open AI, considering its big equity stake in it. But it's turning Microsoft from being just a consumer of open AIs models to a more complex collaboration and also competition.

So Microsoft's internal AI lab is now vying to match or exceed open AIs offering, signaling a future where Microsoft might run its cloud AI services on Microsoft made models instead of open AIs. So in the bigger picture, this development indicates a maturing AI ecosystem. So for users, Microsoft's dual track approach could, if it happens, it could mean more diversity in AI features and possibly lower cost or improved performance, as Microsoft could choose the best or the most efficient model for a given task, whether it's open AIs, its own competitor or a blend of those. So I think in the end, this is a win for consumers.

I think it's a smart move from Microsoft, not having to be, you know, overly reliance on Open AI. I don't think they're going to be dropping Open AI's models anytime soon because they're the best in the world, right? I don't care what anyone says. Go look at the benchmarks.

Go look at Elo's scores, right? Elo scores are, you know, essentially a human puts in a prompt. They get two outputs. They're blind. They don't know which one's which. They say this one's better. Open AI's GPT-4-5 is leading that discussion. So Open AI is not giving up.

It's, you know, seat atop the AI kind of King of the Hill competition anytime soon. And I think if nothing else, this might just push Open AI to continue to improve its AI models at an even faster pace, right, when they may not have as much support and as much data coming in. from Microsoft's integration into Microsoft 365 copilot. All right. So that is a wrap, y'all. So let's go ahead and quickly recap the top stories for the AI news that matters for the week of March 10th. All right. So first, Microsoft has unveiled AI sales agents to challenge Salesforce.

Super Intelligence Startup Reflection AI has launched and come up. out of stealth with $130 million in funding to work on superintelligence. China, there's huge, huge advancements coming out of China with AI startup. Monica has unveiled Manus, which I think could really compete with open AI's products on a lot of different ways. Meta is predicting that hundreds of millions of businesses will adopt its AI agents. Anthropic has urged the creation. of classified info-sharing channels for AI labs with the US government.

Google has unveiled its new AI mode in search, which could signal a big change on how we use the web. Alibaba's QWQ32B model is topping the open source charts. So pretty impressive work there from Alibaba. ChatDB, kind of a low-key update, has given the desktop app on Macs, two-way coding integration, essentially turning it into an IDE or an integrated development environment or at least allows it to better work with popular IDE's. According to reports, Open AI is planning a $20,000 a month professional AI agent.

And last but not least, Microsoft is exploring life beyond just working with OpenAI as it continues to build impressive in-house AI models and also test-true. drive some of Open AI's competitors to see how it'll do powering co-pilot. All right. I hope this was helpful, y'all. If so, don't keep this to yourself. I hope, you know, our shows, especially the Monday show, but all week. I hope that this helps you become the smartest person in AI at your company or your department.

If so, please don't just keep this as your secret. Share. All right. So maybe if you're listening on the Twitter machine, click that retweet or rex or whatever it's called nowadays. If you're listening on LinkedIn, thank you. Tag a friend.

Click that repost button. I'd super appreciate it. If you're on the podcast, thanks for tuning in. As always, I know these shows get a little long. Listen to me on 2X. I won't get mad.

But also please subscribe. Follow the show on Spotify, Apple Music, wherever you're listening, get your podcast. Please leave us a rating as well. Thank you for tuning in. I'll see you back tomorrow and every day for more everyday AI. Thanks y'all.

Meet Firefly AI Assistant. Now live in Adobe Firefly, the Allman One Creative AI Studio. Just describe what you want to create in your own words and the assistant handles the rest, orchestrating multi-step workflows across Adobe Creative Cloud apps, including Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome while the assistant accelerates execution. Stand control with the ability to step in and refine at any time. See it today at Firefly.adobie.com.

And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic, visit Your EverydayAI.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.