The AI Daily Brief: Artificial Intelligence News and Analysis - Why Microsoft's CEO Doesn't Believe in AGI

Episode Date: February 21, 2025

At least...not in the same way others think of it. Microsoft CEO Satya Nadella just sat down for a rare in-depth interview, sharing his thoughts on AGI, Microsoft’s partnership with OpenAI, and the risks of overbuilding AI infrastructure. He argues that AGI should be measured by real economic growth, not benchmarks, and warns that businesses may be overinvesting in AI data centers. Is Microsoft pulling back while others go all in?Brought to you by:KPMG – Go to ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠www.kpmg.us/ai⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ to learn more about how KPMG can help you drive value with our AI solutions.Vanta - Simplify compliance - ⁠⁠⁠⁠⁠⁠⁠https://vanta.com/nlwThe Agent Readiness Audit from Superintelligent - Go to https://besuper.ai/ to request your company's agent readiness score.The AI Daily Brief helps you understand the most important news and discussions in AI. Subscribe to the podcast version of The AI Daily Brief wherever you listen: https://pod.link/1680633614Subscribe to the newsletter: https://aidailybrief.beehiiv.com/Join our Discord: https://bit.ly/aibreakdown

Today on the AI Daily Brief, why Microsoft CEO Sachi Nadela thinks differently about AGI. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show notes. Hello, friends, we definitely have something a little bit different today. Microsoft CEO Satchinadella just gave a really interesting interview about a wide-ranging set of topics. And given his prominence in the AI space, it has caused a ton of chatter. So what we're going to do today is dig into that interview, give the high, highlights, talk about the discussion around it. And in fact, that will be the main episode.

We'll be back with our normal format of headlines and main episode tomorrow. But for now, let's listen to what he has to say. Welcome back to the AI Daily Brief. It's not often that we have another podcast as the context for an entire main episode, but this one, I think you'll see absolutely deserves it. For those of you who don't know, the Dwar Keshe podcast has emerged over the last couple years as the new thinking person's interview podcast. It's in the long format interview style of a Rogan or Lex Friedman, but with extremely high-level, extremely intelligent guests,

and with Dwarkeh Patel, the host, having done a ton of research. Now, D'Rkech is really young. He's only a couple years out of school, but he's built a name for himself by just doing way more background research and frankly just giving way more of a crap than other podcast hosts do. He's also very, very smart and can engage with people on a really intellectual level. Anyways, there's a lot to love about the Dwar Keshe podcast.

It's a great example of why my main shows are not interviews. It's a totally different type of format that really can be done excellently well. Earlier this week, Dwar Keshe posted kids don't underestimate the power of a cold email. And it shows the background story of how he nabbed an interview with Microsoft CEO, Satchianadella. Now, Nadella is obviously one of the key figures in the AI space by virtue of his leadership of Microsoft, by virtue of Microsoft's partnership with OpenAI. And this could have been a really basic banal, say the same things you've always said kind of interview.

And boy, was it not. Media has been reporting this like news because Satya gets into everything from AGI to a bubble in CAPEX and lots more. So what we're going to do today is actually dig entirely into the highlights from this interview. Of course, I highly recommend that you check out the whole thing. I will include a link to it in the show notes.

Now, part of the context for this, and the reason it seems like they agreed, other than Sotja just being a fan of Dwar Cash, is that there were a couple of big announcements from Microsoft, specifically around some quantum computing advances, and that created the news context or reason for the interview. But really, over the course of the one hour, it was a much more big pattern trend type of conversation than just trying to push those two announcements. Now, easily the comments that are getting the most attention in online discourse is around AGI. Farsatia, the test of AGI, will not be found in benchmarks or some arbitrary level of task replacement, but is instead about global growth. He said, us self-claiming some AGI milestone that's just nonsensical

benchmark hacking to me. The real benchmark is the world growing at 10%. This deserves some background for those of you who aren't familiar with growth statistics. Growth rates in the developed world are trending at around 2% or basically zero on an inflation-adjusted basis. And so the idea that we would have to see 10% growth from AGI means a total shift in how the global pattern has gone. He drew the comparison explicitly between the AI revolution and the industrial revolution, noting, once again, that the real marker of success is not about hitting some random benchmark, but about making a paradigm shifting impact on the global economy. In that vein, he argued that the real beneficiaries of AI in the long run will not be the tech companies providing it, but instead,

all the firms that harness AI to drive massive productivity gains. Towards the end of the podcast, the conversation once again circle back to AGI. Dwar Keshe trying to nail down whether Nadella believes that AGI will be achievable. Putting his own terms on it, Dwar Keshe defined AGI as, quote, a thing which automates all cognitive labor, like anything anybody can do on a computer.

Nadella responded, this is where I have a problem with the definitions of how people talk about it. Cognitive labor is not a static thing. There is cognitive labor today. If I have an inbox that is managing all my agents, is that new cognitive labor? Nadella claimed that today's cognitive labor

could be automated, but flagged there would be new cognitive labor that arises from the process. He continued, that's why I make this distinction, at least in my head. Don't conflate knowledge worker with knowledge work. The knowledge work of today could probably be automated. Who said my life's goal is to triage my email? Let an AI agent triage my email. That level of agent, Nadella proposed, would free up the person for a higher level of cognitive labor,

but not eliminate their role entirely. Do Arkesh then suppose that AGI could get to that second level of doing the higher-order cognitive tasks? Nadella claimed then there would be a third layer, adding, why are we thinking that somehow, when we've dealt with tools that have changed what cognitive labor is in history, why are we worried that all cognitive labor will go away? He drew on the Herbert Simon point that humans exist in a state of bounded rationality. That is, no person has the bandwidth to fully optimize their experience.

Instead, they settle for an adequate outcome. Such as supposed, if the bounded rationality of humans can actually be dealt with because there is a cognitive amplifier outside, that's great. Now, let me take a pause here. This is something I've talked a lot about on this show and on other people's podcast as well. My strong sense of what the pattern is going to be, hold aside AGI, just as agents get more profligate and more able to do more of the things that we do today. I think that inevitably there will be a phase one where companies cut a huge amount of human

labor because the productivity gains that they're going for, quote unquote, are just in the form of doing the same stuff they do now, but cheaper. Part of why enterprises are so attracted to agents is that their ROI is so implicit. they work, they do tasks that humans do now for cheaper. We know how markets react. We know that Wall Street thinks in terms of quarters, not years and decades. And so, of course, companies that are able to have the same output with half the number of inputs from a cost perspective are going to be rewarded, at least in the short term. However, I also believe that those companies will not be the ones who write the story of the future. If you do, save half of your budget, to get the same

output, there is no one deterministic way in which you have to apply those savings. Sure, you could just buy back stock, as some will do and play the financial engineering game, but others will reinvest those savings in transforming what they actually do, innovating and offering new types of products, offering a higher level of service that wasn't possible before. And this, I believe, is where you'll start to see a different relationship between human and agents in the hybrid workforce. This is when I think you'll start to see companies experiment with what it looks like to make. To make a different relationship. everyone a manager of a team of agents. To use Satya's terminology, the cognitive labor for everyone will shift into a more managerial type of experience where everyone has access to a small

workforce or, frankly, army of agents to do whatever they had been doing before, but at a much higher scale. Obviously, this is not going to be an easy transition. It requires a totally different mindset, a totally different skill set, and there's going to be fits and starts and challenges along the way. But the point is that I simply do not believe that the replacement of all of what is currently cognitive labor or knowledge work will mean that there is no cognitive labor or knowledge work left. The only question is how long the transition takes and how fast companies can expand their field of view to have big enough ambitions to take advantage of the new capacity that people have when they are paired with teams of agents. Now, bringing it back to this interview, the notion that

AI is an amplification tool was threaded throughout the conversation. And indeed, this question of agents was discussed throughout the conversation. Satya views AI enhancement similarly to the shift brought about by personal computers and communications throughout the 1980s and 1990s. Basically, he pointed out that these advances didn't simply update existing workflows but required fundamental redesigns in workflow. He said that is what needs to happen with AI being introduced into knowledge work. When we think about all these agents, the fundamental thing is there's new work and workflow,

a la what I was just saying. By way of example, he discussed preparing for this podcast appearance. He used co-pilot to gather the documents he would need to familiarize himself with, generate a summary, and even put it in a podcast format. He then shared those resources with his team. Summing up, he said, so the new workflow for me is I think with AI and work with my colleagues. Another comparison that he made was the Lean process being introduced in manufacturing. Lean was first introduced by Toyota in the post-war period and involved running factories with a minimum stock of parts and only finishing enough goods to fulfill orders. It later evolved into what became known as just-in-time manufacturing and focused on cutting out all

waste and bottlenecks within the process. Satchi said that's what's going to come to knowledge. This is like lean for knowledge work. Adding some heft to my idea that everyone's going to have armies of agents running around, Satcha explained that he now has around 10 agents working on small tasks for him, things like sorting email and drafting responses. And so he's shifted. Instead of a bloated inbox waiting for him every morning, he now has a list of agentic tasks waiting for approval.

Nadella commented, I feel like there's a new inbox that's going to get created, which is my millions of agents that I'm working with, which will have to invoke some expectations for me notifications for me, ask for instructions. And it definitely feels like this is a problem that Microsoft is aiming to solve. Satcha continued, what I'm thinking is that there's a new scaffolding which is the agent manager. It's not just a chat interface. I need a smarter thing than a chat interface to manage all the agents in their dialogue. That's why I think of this co-pilot as the UI for AI is a big deal. Each of us is going to have it. So basically think of it as there is knowledge work and there's a knowledge worker. The knowledge work may be done by many, many agents, but you still have a

knowledge worker who is dealing with all the knowledge workers. That's the interface that one has to build. We could have a whole separate conversation and maybe this is a podcast that I'll do at some point about what I see or frankly what we're seeing across super intelligent as the new infrastructure for agents and for the hybrid workforce that they implicate that needs to be built. But this sort of agent management tool is part of what will help people transition from thinking about themselves as the end worker who does things to the manager who coordinates things. Now, as big and different as vision for the future is. Nadella also pointed out that we are all vastly underestimating, effectively the human friction that's going to be a part of this transformation. For example,

he said that supposing that AI can automate something like 60% of global work, he said, I think that in order to have a stable social structure and democracy's function, you can't just have a return on capital and no return on labor. We can talk about it, but that 60% has to be revalued. In my own simple way, maybe you can call it naive, I think we'll start valuing different types of human labor. What is today considered high value human labor may be a commodity. There may be new things that we will value. He gave the example of care and nursing work as something that could become highly valued, adding, ultimately, if we don't have a return on labor and there's meaning and work and dignity and work and all of that, that's another rate limiter to any of these things

being deployed. The other big rate limiter that Nadella brought up was the legal system. We're talking about all the compute infrastructure, but how does the legal infrastructure evolve to deal with this? The entire world is constructed with things like humans owning property, having rights and being liable. If humans are going to delegate more authority to these things, then how does that structure evolve? Until that really gets resolved, I don't think just talking about the tech capability is going to happen. He actually connected the dots between the legal challenge and the alignment challenge. He said, you cannot deploy these intelligences until and unless there's someone indemnifying it as a human. This AI takeoff problem may be a real problem, but before it is a real problem, the real problem will be in the courts. No society is going to allow for

some human to say, AI did that. And indeed, this was another one of the big themes of the conversation, that AI deployment doesn't happen in a vacuum. Basically, Satya fundamentally doesn't believe advanced AI will suddenly come into existence at proliferate unchecked. He pointed out that society has already developed software capable of producing extreme harm, but guardrails are developed to mitigating the risks. We don't just write software and then just let it go, he said. You have software and then you monitor it for cyber attacks, you monitor it for fault injections, and what have you. Today's episode is brought to you by Vanta. Trust isn't just earned, it's demanded.

Whether you're a startup founder navigating your first audit or a seasoned security professional scaling your GRC program, proving your commitment to security has never been more critical or more complex. That's where Vanta comes in. Businesses use Vanta to establish trust by automating compliance needs across over 35 frameworks like SOC2 and ISO-27-01. centralized security workflows, complete questionnaires up to 5x faster, and proactively manage vendor risk. Vanta can help you start or scale up your security program by connecting you with auditors and experts to conduct your audit and set up your security program quickly.

Plus, with automation and AI throughout the platform, Vanta gives you time back, so you can focus on building your company. Join over 9,000 global companies like Atlassian, Cora, and Factory who use Vanta to manage risk and prove security in real time. For a limited time, this audience gets $1,000 off Vanta at vanta.com slash NLW. That's V-A-N-T-A.com slash N-L-W for $1,000 off. If there is one thing that's clear about AI in 2025, it's that the agents are coming. Vertical agents by industry, horizontal agent platforms, agents per function. If you are running a large enterprise, you will be experimenting with agents next year. year. And given how new this is, all of us are going to be back in pilot mode. That's why Super

Intelligent is offering a new product for the beginning of this year. It's an agent readiness and opportunity audit. Over the course of a couple quick weeks, we dig in with your team to understand what type of agents make sense for you to test, what type of infrastructure support you need to be ready, and to ultimately come away with a set of actionable recommendations that get you prepared to figure out how agents can transform your business. If you are interested in the agent readiness and opportunity audit, reach out directly to me, NLW at B-Super.aI, put the word agent in the subject line so I know what you're talking about, and let's have you be a leader in the most dynamic part of the AI market.

Hey listeners, want to supercharge your business with AI? In our fast-paced world, having a solid AI plan can make all the difference. Enabling organizations to create new value, grow, and stay ahead of the competition is what it's all about. KPMG is here to help you create an AI strategy that really works. Don't wait. Now is the time to get ahead. Check out real stories from KPMG of how AI is driving success with its clients at KPMG.us slash AI. Again, that's www.kpmg.com.com slash AI. Now, back to the show. All right, so we have this whole big section in parts of this conversation that were about AGI and agents in the workforce, the newly hybrid workforce. What sort of barriers that aren't

technological will slow down the rate of change. Maybe the other most discussed part of the interview, though, was around capital expenditure and the risk that data centers will be overbuilt. And interestingly, and this is what got people really chattering, Nadella acknowledged that the infrastructure will be overbuilt, just like it was during the dot-com era and even going right back to the construction of railways. He commented, I'm so excited to be a leaser because I build a lot, I lease a lot. I'm thrilled that I'm going to be leasing a lot of capacity in 2027, 2028, because I look at the builds and I'm saying, this is fantastic. The only thing that's going to happen with all the compute builds as the prices are going to come down. He brought in his experience during the cloud

computing buildout as the industry transitioned away from on-premise servers and said, once we started putting servers in the cloud, suddenly people started consuming more because they could buy it cheaper and it was elastic and they could buy it as a meter versus a license and it completely expanded. With the deep seek news cycle still front of mind, Nadella made the point that training a model is only half the battle. He said it's not about building compute. It's about building compute that can actually help me not only train the next big model, but also serve the next big model. Until you do those two things, you're not going to be able to really be in a position to take advantage of your investment. So that's kind of where it's not just a race

to building a model, it's a race to creating a commodity that is getting used in the world. You have to have a complete thought, not just one thing you're thinking about. Now, with all of this, when Dwar Keshe suggested it might be more logical for Microsoft to increase their capex to 800 billion, rather than 80 billion if they believe the opportunity is so vast, Nadella said, the classic supply side is, hey, let me build it and they'll come. That's an argument, and after we've all done that, we've taken enough risk to go do it. But at some point, the supply and demand have to map. That's why I'm tracking both sides of it. You can go off the rails completely when you're hyping yourself at the supply side versus really understanding

how to translate that into real value for customers. The interpretations for this were fairly dramatic. Prakash the Adepai account writes, Satya is out. TLDR and Microsoft doesn't believe in AGI, AGI, wary of overinvestment, open AI partnership is over. Adipai calls out, very negative on CAPEX spend from Microsoft. Quote, if you look at the Industrial Revolution, there was a lot of money lost. Countries are going to deploy capital. I'm so excited to be a leaser. I build a lot, I lease a lot.

Adipai continues opinion. Satya is calling the bubble a buildout. The crazy people like governments are entering the game. He's happy to lease from them when they overbuild. His own CAPX spend is capped. Lynn Alden writes, Microsoft pushing hard on quantum and then dunking on AGI Kappex kind of put Sotia as the main character for a while. I think the way that people interpreted this Cappex conversation is interesting.

I'm not so sure it's as dramatic as people are making it out, but a lot of people had a similar reaction to this. Ultimately, another big conclusion was Nadella's belief that AI won't be a winner-take-all market. He recalled comments during the cloud buildout that AWS had an insurmountable lead, rendering Microsoft Azure as a pointless endeavor. Nadella said, consumer markets can sometimes be winner-take-all, but anything where the buyer is a corporation and enterprise and IT department, they will want multiple support. and so you've got to be one of the multiple suppliers.

He extended this thought to the market for models. His belief seems to be that open source will act as a governor on closed source models, much like Linux successfully checked Windows during the 90s and 2000s. Regarding the commoditization of AI and the lack of moats, Nadella commented at scale, nothing is commodity. He pointed out that every tech company in the world was capable of racking servers, meaning that cloud storage was definitely a commodity, but that the real moat was in hyperscaleing.

In other words, spinning up a cloud services business was trivial, But serving cloud storage globally at high speeds with minimal downtime was an error-defining challenge. He seems to think a similar paradigm will play out in AI, stating, in the model layer, models need ultimately to run on some hyperscale compute. So that nexus is going to be there forever. It's not just the model, the model needs state. That means it needs storage, it needs regular compute for running these agents and the agent environments.

Ultimately, he made it clear that the AI race in his mind is not nearly as simple as it's being perceived. You have to not only get the tech trend right, you also have to get where the value is going to be created with that trend. These business model shifts are probably tougher than even the tech trends. I think part of what makes this moment so interesting is that the deeper you get into the AI space, the more convinced you are that we're still underestimating the magnitude of the potential disruption. I think Sotia is correct to point out that there's going to be non-technology-based

delimiters, societal barriers, legal barriers, human inertia, in other words, that slows down what the technology could do. But we still are talking about a fundamental shift. where the work in a decade really just does not look like the work today. It's almost impossible for most people to actually understand or grok just how big of a difference that's going to be. And that's why these types of conversations are so valuable. They help us understand at least the thoughts of some of the leaders in the space who are pushing things forward. Anyway, great conversation, highly worth the time to go check it out. For now, that is going to do it for today's AI Daily Brief. Until next time, peace.