231 | OpenAI’s Dev Day is showing ambitions for taking over the world, Anthropic’s mega deals with IBM & Deloitte, AI bubble fears vs. fundamentals, and how AI is reshaping the economy - AI news for week ending on October 10, 2025

0:00Hello and welcome to a weekend news episode of the Leveraging AI podcast, the podcast that shares practical ethical ways to leverage AI to improve efficiency, grow your business, grow your business, maybe a mix of both. And we are going to cover some interesting new data points when it comes to the impact of AI on jobs, the impact of AI on the economy. The first section in which we're going to talk about OpenAI and DevDay and its implications is actually a recording from our AI Friday Hangouts, which is an open to the public session that I'm doing every single Friday at 1 p.m. Eastern. It's a community of people who care about AI and how it's going to impact their lives and their businesses.

0:59We get together every single Friday at 1 p.m. and we discuss AI. Sometimes it's more tactical. Sometimes it's more strategic. Sometimes in this case, somebody asking me about what happened in Dev Day and can I give them a quick recap and what I think about it. So this is the section you're going to hear. If you want to join us every Friday, like I said, it's open to the public. Just drop me a note on LinkedIn or send me an email. Tell me that you're interested in joining us. We usually have about 25 to 30 people every single week and we just discuss it. And just me, other people are sharing their findings and things that they have learned and so on.

1:32So it's a great way to keep up to date and learn things about AI when it comes to business. So as I mentioned, just let me know if you want to join. But now let's start with the news.

1:48So OpenAI had their Dev Day this week. They had Dev Day. On Dev Day, they usually don't release necessarily new models, even though they kind of had two announcements that are model related. So the two announcements were a smaller version of their voice engine and a smaller version of their image engine that are delivering, you know, 85 % of the capability at 70 % discount on the tokens. You see that more and more. That's happening a lot recently. you know, the first company we can also do in this is DeepSeek, which didn't start with their model, but they still did the same process. And since then, we've seen this again and again and again with these companies.

2:34The latest one is Grok4 Turbo, whatever they're calling the smaller Grok4, which achieves like 90 % of what Grok4 can do at like an 80 % discount and faster. And like all these models are doing the same thing. So OpenAI came out with the same thing for their voice API and for their image generation API. And so you can now do this, which is going to save you a lot of money and get, for the vast majority of cases, a good enough solution, which is fantastic. So this is kind of like was the very least interesting stuff that they released. Usually Dev Day is around tooling and specifically tooling for developers, right?

3:14That's the whole point. so they've released several different things the most interesting one is apps inside of chat gpt right so if you remember they had plugins way way back and then they sunsetted it because it didn't really work so this is the new version of that but it's a lot more interesting in the way they've implemented it so the the idea here is first of all a lot of the infrastructure is already there because behind the scenes, it's actually using MCP. So the implementation of MCP allows you to bring any third-party software and connect with it in a very seamless implementation. And then the idea is that the application can...

3:59So it doesn't take over the chat, but the chat is aware of what's happening in the application, right? And so you can run a search for travel and it will show you like a travel box, but then you can go and ask questions. Oh, I'm actually looking for a four-star only. And it's, oh, okay. And then it's going to do the thing. So it's aware of what's happening. They've shown Khan Academy and how you ask for a lesson, it's going to do the thing. And then you can ask follow-up questions that do not appear in the lesson in Khan Academy. But because Chachupiti is aware of what's happening in the Khan Academy lesson, now Chachupiti takes over and gives you additional guidance or explanations and so on.

4:41So that's one great example. And they created an AppSats SDK that any monkey can now create these apps for OpenAI. So in the release, they had, I don't know, like 20 launch partners, and they already said that there's another 20 in the queue. But the idea is now that there's AppSats SDK, anyone can put their app on ChatGPT. this if it works this time and it doesn't matter if it's this time or the next time around eventually it will work which means it's taking over the internet right this is replacing the browser combine that with slightly more agentic capabilities that will allow on its own to use the apps without you having to figure out which apps to use and it has its tooling becomes everything.

5:39And so this is a step in the direction of A, open AI taking over the world, B, an agentic universe that has access to an endless number of tools at its fingertips and it can pick and choose which one to use for what purpose. so if we take the trip example right now you have to go through the connector and say look for all the emails that I have about my upcoming travel and then it will find the emails and you copy that and say okay for this trip from blah blah blah I won't but in the future it will just say okay I want you to take care of my travel for the next two weeks based on the emails that I and the things and the agreements I've signed and it will it will go and find the emails and look for the thing and we'll do the decluttering of the information it will figure out the right flights and how long you want to stay and if it has your context going back to our original conversation somewhere knowing that you like to stay at Marriott because you collect their points and you like flying red eyes back from the west coast and you don't like flying with spirit for obvious reasons and so on and so forth uh then it will do the thing.

6:56And if it's really nice, it's like, okay, here are two options. Which one do you prefer? That's where it's going. And to be honest, that even won't require you triggering it because it will be aware of what you're supposed to do. And it will say, hey, I'm booking your travel for this thing. Is that okay? You won't even have to say anything because you just signed an agreement that you have to be at so-and-so place for these two days or something. like meeting, training, sales, delivery, like whatever the case is. It will be aware of that because it has access to your inbox. So it's just, from OpenAI's perspective, when I mean taking over the world, you heard me say that a thousand times, I'll say it a thousand more.

7:43Context, context, context, context. It's just more context. They will know, think about Google. Why is Google one of the largest companies in the world? because it's into everything we do. It controls our browser. It controls our search. It controls our phones. It controls our videos. They just know more stuff about you. OpenAI are competing on the same thing. And so this is one thing that they've done is the apps. The second thing that they've done is the whole agent kit, which has several different components, one of them is an agent builder and then there's different overlays on top of the agent builder that is supposed to help you evaluate and make it safer uh kind of like on its own and this is you know a shot right into the gut of n8n make.com zapier lindy like all all of these uh and And it's a step in the right direction.

8:49Like you see a lot of people saying, yes, this is going to kill Any10 and it's going to kill Zapier and so on and so forth. And the answer, at least of right now, is not yet. Whether that's actually going to happen or not, time will tell. It is very, very hard to compete with 800 million weekly users. it's very very hard to compete with i want to raise another billion dollars so i'll do it tomorrow when when when na10 just did a huge announcement that there is 180 million dollars which two years ago would have been holy shit that's incredible round c uh but now that you're competing against open ai that's a whole different story that doesn't sound like a lot of money.

9:42The other crazy thing about this whole agent thing, which again is not at the level of NA10 yet, they're claiming that they've used OpenAI codecs and a very small team to develop the whole thing from idea to deployment in six weeks. That is insane. Now, yes, it's not at the level of Zapier or anything, whatever, but it's a nice step in the right direction in six weeks, which tells you how crazy good these AI development tools are right now. now whether they're using a specialized version of codex or like the general available it doesn't matter because even if there is using they are using specialized one that's going to be the next version they release so it's like uh the the timelines here are very very short and so being able to release a agentic building tool that raises the question whether it can compete with NA10 and Make that has been around for a long time tells you how crazy the software world is right now.

11:04And then the last thing that I will say about what are the gaps that it has now that I think are not major. One is connectors to existing tools. So, you know, NA10 has all the existing nodes. Make has even more. Zapier has even more. I don't see that as a big gap, especially in the world of MCP. So if you don't have to sit somebody down to write the API to this tool because you can just connect to the MCP, then the timeline to connect to some of these tools become, I'll add a tool every 10 minutes, assuming the MCP was done properly, which not a given. I'll give you an example in a minute. So that's number one.

11:54Number two is there's a lot of stuff in Zapier and A10 and so on that are not agentic. It's just a regular automation flow, which has huge benefits. The biggest benefit, there's no statistical question there. It will do the same thing every single time. And an agent won't. that's the benefit of an agent that it doesn't do the same thing every single time. It actually thinks about what it needs to do. But there are many cases that you want to mix and match these two things. You want a process that will go step by step exactly the same way every time and then think about it in one or two of the steps.

12:41It doesn't do that at all. Now, will it do that in the future? I don't know. If it takes six weeks to do, probably yes in six weeks. but I don't know that. And so I really think that this is a, if I've learned anything from this Dev Day, and it was very obvious before, but this made it like crystal clear in the open, we are here to take over the world every time, everywhere, and everything across anything that anybody does. that's basically if you want to sum up what they've announced this is what they've announced we don't care what you do we're coming for you uh

13:31which you know for a company that nobody knew that existed uh in 2022 and is now worth whatever whatever they decide because that number keeps going up every few weeks, as of right now, half a trillion dollars, and can raise whatever amount they want in whatever pace they want, and has the most talented people and unlimited compute and unlimited capital. I don't know if I want to bet against them. And now from OpenAI to Anthropic, Anthropic also had a very meaningful week. They have signed two very significant deals. But before we discuss the deals that they've signed, let's talk about something very interesting that they have released or announced this week.

14:20So they just launched two public beta plugins for their development environment. So this adds slash commands to their development solution that allows to integrate several different capabilities. And these come in four different customization types, plugin bundles slash commands, which allows to provide custom shortcuts and additional capabilities, sub-agents, which are specialized task agents that can do specific things within your coding environment, MCP servers, which add tools and data connectors to your environment, and hooks, which are adding workflow behavior modifiers. Now, the idea is that these plugins enable developers to package and share complex setups like debugging workflows and deployment pipelines with just a single command instead of an entire setup that was required before, which means developers inside a company or across open source universes or just sharing with the world can share a lot of the stuff that they're doing and enhance the coding environment very quickly and very easily.

15:22Now, in addition, to make it even easier to share, anyone can host a plugin marketplace via a Git repository or just a URL. And the idea here is obviously to create a flourishing and fostering community that will develop and share these kind of plugins and additional capabilities for cloud code. Now, examples that they gave is that these plugins can be used to support enforcing team standards, aiding open source contributors, sharing productivity workflows, connecting internal tools, and bundling framework-specific customizations. Out of the box, Anthropic are sharing four different things. They are sharing plugins for PR reviews, security guidance, Cloud Agent SDK development, and Meta plugin for creating new plugins.

16:06I anticipate this thing to catch like wildfire and be something similar to MCP. This might become a new standard on how AI tools are sharing advanced and more customized capabilities. Definitely within companies, I think this will flourish, but I also see like an international open source community coming around this because this provides a lot of value. But as I mentioned, in addition to this, Anthropic signed two mega deals this week. The first one we're going to discuss is their deal with Deloitte. Deloitte is going to be providing cloud access to over 470 ,000 professionals that work for it all around the world.

16:46In addition, they're going to work together to develop what they're calling the Cloud Center of Excellence, which Deloitte will establish together with Anthropic in order to train experts on how to implement and develop frameworks and solutions using different cloud tools. They have already created a certification program for 15 ,000 Deloitte professionals that will become the champions for cloud implementation across the entire organization. This is a very similar approach to what I do with most of the companies that I work with, where we establish a center of excellence, we establish an AI committee, and we create a network of specific individuals who either raise their hand or the organization thinks are the best fit to become the champions across the different departments.

17:29What this does is it guarantees that in addition to the top-down approach, you're getting a bottom-up grassroots implementation and innovation across the entire company. Now, in addition to obviously internal work, the goal of this partnership is to create tools and solutions tailored for the clients of Deloitte across multiple industries with a specific focus on regulated industries such as financial services, healthcare, public services, which all prioritize compliance, transparency, and security, which is exactly the things that Anthropic are known for. So again, a huge deal for Anthropic, the biggest deal they signed so far from the size of the enterprise client that they're signing.

18:06Another really big deal that they signed this week is with IBM. In the partnership with IBM that was announced on October 7th, IBM is going to integrate Claude's large language models into their IBM portfolio, starting with the AI IDE that they're developing jointly. Claude will practically be the AI first IDE, also known as Integrated Development Environment, which is the platform that developers use to write code. And it is already available for private preview for select IBM clients. It is automating tasks like code modernization, upgrades, migrations, refactoring of code, and so on across the entire software development lifecycle.

18:47And as we just heard, the additional new capabilities to create plugins and so on, all of that is going to become a part of what IBM is offering their clients. In addition, in early testing internally in IBM with over 6 ,000 IBM internal users showed an average productivity gain of 45 % with cloud generating tasks, code suggestions, and handling security compliance checks, 45 % is a very significant number. It means that those 6 ,000 people more or less are working twice as fast than anybody before that inside of IBM. And if they can sustain that and obviously deliver that to IBM clients, that is very significant.

19:27Now, IBM is also exploring integrating Claude into some additional software products, including their Watson Assistant. They also work together to co-develop what they're calling architecting secure enterprise AI agent with MCP, which is a guide that is outlining the agent development lifecycle, ADLC, similar to the concept of SDLC from the software world, for building, deploying, and securing enterprise-grade AI agents with open standards. This, I believe, is very good news. I think it's a great partnership. It provides the stability and the reach of IBM, together with the innovation and the new AI capabilities of Anthropic, to a very large audience.

20:06Or like Dinesh Nirmal, who is the senior vice president of software at IBM said, this partnership is giving development teams AI that fits how enterprises work, not experimental tools that create new risks. And I couldn't have said it better. I am certain that IBM clients will be excited to receive these capabilities. IBM investors definitely liked it as their shares rose 7.5 % in early trading post-announcement. Speaking of jumping stock prices, there is a growing chatter on the fact that we might be in a very serious AI bubble, even though they are people who are thinking the other way around.

20:42But the voices who are claiming that the bubble is very clear are growing stronger. In a very interesting article from Axios this week, they shared several of such opinions. And the main thing that they're highlighting is off-balance sheet debt that is being used in order to finance some of the crazy investments that we're seeing in the past few months, which feels very similar to what was happening in the dot-com crash with crazy debt funding that led to eventually the collapse of that very big hype, and very similar to the 2007-2008 financial meltdown where different, not totally clear financial tricks were used in order to finance the growth with a lot of debt that was raised in really weird ways.

21:30Well, apparently there is a growing phenomena of corporate debt that is quietly ballooning in ways that are not the standard ways of raising money, which means ways that are less visible to the shareholders. Two examples that were raised in this article, Meta is raising$29 billion in private capital for AI data centers via SPVs, which is allowing the company to avoid reflecting that debt on its balance sheet. A similar approach is being used by Oracle that has issued$18 billion in bonds in order to bankroll its AI infrastructure expansion. Very similar kind of thing. Or as Dario Perkins, the managing director of Global Macro at TS Lombard, which is a global data and investment intelligence company, said, and I'm quoting, it feels much closer to 2000 than 1995.

22:19And he continued, I wouldn't touch this stuff now. What basically he's saying is that what's happening behind the scenes with the amount of kind of like shady debt that is being raised in order to finance some of these things feels a lot more like the end of the dot-com boom versus the beginning or the middle of it. Investor and author Paul Kedrowski said, and I'm quoting, the market is rewarding them even if it makes no economic sense to spend at this level because there is no way they can recoup the value of the capital spending. So that's another viewpoint of somebody who thinks we are definitely heading into a big bubble.

22:56In addition to these death mechanisms, Axios also mentions investment recycling and insider selling basically means people from the inside who started selling their stock because they don't necessarily believe it's going to go higher, or maybe they know something about where it's going next, which might be the opposite direction. All of these are very obvious red flags. What exactly is investment recycling? We're going to talk about a very interesting example later on in this episode. Two other very prominent figures who shared their concerns is Jeff Bezos and David Solomon, who is the CEO of Goldman Sachs.

23:29Well, Bezos said, and I'm quoting, this is a kind of industrial bubble talking at the Italian Tech Week. And he basically says that investors are currently struggling distinguishing between good ideas and bad ideas because of the amount of hype that there's around it. So there's a lot of money going not necessarily in the right direction. That being said, he was very clear that he believes AI has a transformative impact across industries and across different sizes of companies and across the entire globe. So he's not saying that AI will not transform everything that we know. He's just saying that some and maybe many of the investments are bad investments at this particular point and that many people don't know how to tell between good or bad.

24:09And Solomon said, and I'm quoting, I wouldn't be surprised if in the next 12 to 24 months, we see a drawdown with respect to equity markets. Again, that's coming from the CEO of Goldman Sachs. He knows one or two things about making these kind of predictions, and he wouldn't make them unless he actually believes this is where it's going. That being said, he's talking about 12 to 24 months. That's a pretty long timeframe. And to give you an opposing opinion, Dan Ives from Wedbush Securities argues that the AI market that is underlined by huge investment in hardware, real investment in actual goods and in huge growth in services in companies such as ChatGPT that is seeing real growth and real demand defers and is dramatically different than the dot-com bubble.

24:53And he's predicting a two or three years tech bull run with accelerating enterprise spending and implementation. So you can see views to two different directions, but here are a few other data points for you. There was another great article this week by The Information. The Information is a great source of internal reliable information from the leading companies in the world. And they shared about the very thin margins that Oracle is experiencing in this crazy expansion that they're going through right now. So we shared with you some of the latest deals by Oracle. Their stock price have jumped through the roof on the news of the new deals that they signed with projected revenue of$381 billion in the next five years.

25:34However, internal documents are revealing that there's a razor-thin margin to this revenue, which is raising concerns and questions about, A, the sustainability, and B, the level of risk that they're taking in order to do this. So Oracle Cloud Business has generated$900 million in revenue in the quarter ending on August 2025, but the gross profit of that was only$125 million, which is 14%. The profit margin of their AI cloud server rentals has ranged between 10 % and 20%, averaging 16 % in the past year. Now, without any benchmark, that is meaningless. But the average profit for overall Oracle before this AI expansion was 70%, 7-0.

26:20So 14 to 16 is a very, very small margin compared to what they are used to. Add to that the fact that Oracle lost nearly$100 million on NVIDIA Blackwell chips rentals in this past quarter, partially due to the lag between data center setup and customer usage. But it's still showing you that a lot of the stuff that they're doing right now is not necessarily profitable, which means together with the small margins might be a very high risk. Combine that with the fact that some of their margins are further squeezed by the fact that some of the leasing data centers are actually from a third party, meaning they're just the middleman in which they have even smaller margins, which is very different than like AWS, Google, Cloud, etc., which own the vast majority of their facilities.

27:07Now, to add even further risk to that, Oracle's top AI cloud customers are five companies, ByteDance, Meta, XAI, OpenAI, and NVIDIA, and they together account for 80 % of its AI cloud revenue, which is a very serious concentration, which raises a lot of risks. There's a lot of companies who have the 80-20 rule. This is more like the 98-2, where a very small percentage of the clients are holding a very, very large percentage of the revenue. So if the relationship with one or two of them goes south for whatever reason, it puts their entire operation at risk. Now, earlier we mentioned circular investments and the risk aspect of them, and we got an amazing example of that this week.

27:52So this week, OpenAI announced a very interesting deal with AMD. Why is it very interesting? It's very interesting for two different reasons. Reason number one, AMD is NVIDIA's number one competitor. And NVIDIA has been the number one backer from a technology perspective and from a partnership perspective to OpenAI since day one. If you remember, Jensen Huang personally delivered the first GPU of the first model and then of the second model to Sam Altman and OpenAI. And so very long lasting, strong partnership between these companies. So going and signing a really large deal with their number one competitor may not be the best idea in the world.

28:30but putting that aside for a minute this deal is structured in a very interesting way open ai are going to get 10 of amd's stock as part of this deal the deal by the way is for six gigawatts of chips which is a huge amount of compute it's an unheard of amount of compute until the last few months where we've seen similar deals through stargate and through the partnership with NVIDIA. And this is going to deliver this six gigabytes of compute with MI450 series chips from AMD to OpenAI over the next few years. Now, this whole deal, if you think about it, is really, really weird because A, how the hell is OpenAI going to pay for the deal with Oracle and then to the deal with NVIDIA?

29:18And then for this, there is no way they can raise this kind of money. But if you remember, they got 10 % of AMD stock. AMD stock rose 43 % this week with 11 % in a single day just on Wednesday, October 8th as a result of this news. So the amount of money OpenAI made on paper, at least by just getting this 10 % of stock and pushing the stock forward can by itself help them pay for a chunk or maybe all of the immediate investment that they need in order to start paying for this deal. So this is literally money being generated out of thin air with OpenAI getting stock, stock price jumping through the roof.

29:56Now they have the money, they can actually pay for the goods that otherwise they did not have money to pay. This sounds like a great way to blow up a size of a bubble without having actual money behind it. That being said, with the amount of runarounds that Sam is doing around the world right now to build international relationships with actual countries. And his push to develop data centers all around the world is showing, just like we said in the beginning, that they're going for world domination. And they're literally trying to get any piece of compute that they can and secure that ahead for the next few years.