Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Phil pods with Exponential View Founder Azeem Azhar: There is no AI bubble: The enterprise velocity gap & China's efficiency moat
139 Plays21 days ago
In this episode of From the Horse’s Mouth, HFS Research CEO and Chief Analyst Phil Fersht sits down with Azeem Azhar, founder of Exponential View, Bloomberg contributor, and author of Exponential, for a candid conversation about where AI really stands inside the enterprise.
Azeem explains why he believes there is no imminent AI bubble, pointing to enterprise investment, customer spending, and revenue growth as signs of sustained momentum. Together, Phil and Azeem discuss the growing AI velocity gap between individuals and enterprises, why legacy systems and leadership remain major barriers to adoption, and how AI agents, forward-deployed engineers, and organizational agility are reshaping competitive advantage.
They also explore China’s AI progress, enterprise trust in AI, workforce reskilling, and why speed of execution, rather than technology alone, will define tomorrow’s market leaders.
This episode is essential viewing for business leaders, technology executives, consultants, investors, and anyone looking to understand the real impact of AI on enterprise transformation and the future of work.
Chapters 00:00 – Introduction 01:23 – AI in the enterprise today 04:24 – Is there really an AI bubble? 08:47 – The AI velocity gap 14:13 – Culture, talent, and AI adoption 21:21 – Building competitive advantage with AI 30:30 – Lessons from China’s AI ecosystem 37:15 – Why enterprise trust in AI matters 44:05 – The future of work and productivity 52:55 – Closing thoughts
Learn more & connect
- HFS Research: https://www.hfsresearch.com
- The podcast: https://horsesmouthpodcast.com
- Phil Fersht on LinkedIn: https://www.linkedin.com/in/pfersht
- Azeem Azhar’s Exponential View: https://www.exponentialview.com
About From the Horse’s Mouth
From the Horse’s Mouth: Intrepid Conversations with Phil Fersht brings together founders, executives, and contrarian thinkers for unfiltered conversations on the forces reshaping business and technology. Hosted by HFS Research CEO and Chief Analyst, Phil Fersht.
#AI #ArtificialIntelligence #EnterpriseAI #FutureOfWork #AzeemAzhar #PhilFersht #HFSResearch
Recommended
Transcript
Introduction to Azim Azhar and His Work
00:00:02
Speaker
You're listening to From the Horse's Mouth, intrepid conversations with Phil First.
00:00:14
Speaker
welcome to the latest edition of From the Horse's Mouth web podcast. I'm your host, Phil First. And joining me today is somebody I think I've known for about nearly 10 years now.
00:00:25
Speaker
And when I first met him, he came to one of our HFS summits in London and spoke all about AI. This was when I think everyone was obsessing with automation and process. But um Azim's been a regular voice on AI for a very long time. He's the founder of Exponential View, which is one of the world's most influential platforms for understanding the impact of tech on business and society. So if you get a chance, sign up for Exponential
Azim's Return to Coding and Tool Development
00:00:53
Speaker
View. i It's one of my favorite emails when I get it through on a weekly basis. um you know I'll get Azim to talk more about himself, but you know he is everywhere. He's got a very interesting show on Bloomberg TV. He's a frequent speaker and attender at the World Economic Forum, amongst other things.
00:01:14
Speaker
And he's also a fellow at Stanford, Oxford Martin School, and and Harvard Business School. And in 2022, he published his debut book, Exponential Order and Chaos in the Age of Accelerating Technology. So great to have you, Azeem. Have I missed anything?
00:01:30
Speaker
Well, it's great to be back, Phil. Well, you did miss out the detail of a very, very posh and ah seductive hotel that you held that HFS event in back about 10 years ago, which, as you say, was all about ah RPA, if if that's still a
Is There an AI Bubble? Evaluating the Risks
00:01:47
Speaker
thing. And yeah, I mean, it's been a really, really busy 10 years. I actually love the fact that I've been able to ah get back to writing code, which is not something I had done since 2010. And, ten and you know, spend a lot of spend a lot of time doing that and a lot of the tools I use day to day now are things that I have designed myself and had built by Claude Code or something similar.
00:02:11
Speaker
Wow. So you're one of these coders whose life's been transformed by frontier models? Well, I'm but'm not so sure I was a coder because I remember my head of engineering telling me ah back in 2010, please stop doing this, by which he meant write code. And and so I did did, in fact, stop. But actually, what you know what ends up happening is that do you know one knows about software and you know about products because you've been in the industry for a while and you know about your own needs. So now that you can articulate that and and essentially get the code hammered out that you that you need and you can
00:02:53
Speaker
make sure that all those things that you would never have done yourself as a two-finger coder, like yeah refactor it or have good test coverage, or that all gets taken care of. um and And it's been pretty it's been pretty remarkable. and In fact, in the last...
00:03:11
Speaker
in the last eight or nine weeks, ah my interface to writing the things I need has become Discord because I just fired off to my OpenClaw agent and my OpenClaw agent has access to clawed code, it has access to to OpenAI's codecs and it goes off and builds whatever whatever I need. and i And you know interestingly, every night it will go off and walk all of my repos on GitHub and it will fix certain bugs that it finds and it will refactor you know monoblocks of of code. It'll check them for security problems. It'll update libraries if there's been an advisory to do so. And it does all of that in the background. And the it kind of works. I mean, it kind of works. yeah
00:04:05
Speaker
and We'll get onto it later, but why can't enterprises do the same thing as us individuals? But let's let's go back to where we are today. And you know you've been very clear for a while there's no AI bubble.
00:04:19
Speaker
um So you know being over in the States these days, you know all I hear is this fear of the bubble bursting and and and what's going to happen and doom and gloom. there's a lot of that going on.
00:04:31
Speaker
I so don't get so much actually outside of the States. so Why do you think the sceptics are getting it wrong? We did a lot of work to understand what do you need to look out for to see if there's a bubble approaching. i mean, one way to think about that is a bubble bursting is a bit like a heart attack. And anyone can have a heart attack. But if somebody is morbidly obese with really high blood pressure, never exercises and eats a you whipped cream, ah they're more likely to have a heart attack than someone who doesn't have any of those those attributes. In a sense, what are those biomarkers? Because you you can't really predict this. your Bubble bursting is really just about change in in sentiment and ye that could be a ah stampede.
00:05:17
Speaker
So the things that we we looked at were ah ah these these markers, and the markers were how much how much is being invested in in the underlying assets, because bubbles always form when there's too much investment, relative to the size of the economy.
AI's Impact on Enterprises and Revenue Growth
00:05:36
Speaker
How much revenue is there to support the investment in those assets? How fast is revenue growing? Has the market, independently of all of that, lost its marbles, and is it you know valuing a company ah a million times its earnings, a public company. And the is the quality of the funding that is coming in to to finance all of this high quality? Is it low quality? Is it heavily securitized and tied up in ways that you you make you feel uneasy? And we quantify all of those all of those measures. So our framework is completely different.
00:06:12
Speaker
ambivalent as to whether we're right about to at a bubble that's going burst or we're in ah we're in a bust cycle. It doesn't care about any of that. What it does is it measures and it it gives me a sense of what is the risk that we are we are near a bubble. And when we look at that data as we did back in in August last year, we looked at it again at the start of this year, we're just looking at it again now.
00:06:35
Speaker
not the indicators are not read enough to say that there's going to be a sustained bubble bursting. Maybe there might be a correction, but a bubble bursting is, we we you know we remember this, this is the global financial crisis, this is the telecoms bubble, dot com bubble of 2000, 2001. And that's to say that this wave and that's not to say that this ai wave would may not end up with a bubble bursting, but it for in my view, and more likely than not, we've still got time to run because of where the indicators currently sit. And so why do I say that? Well, we know that there is hundreds of billions of dollars being poured into capital expenditure to buy land power, chips, memory, cooling, and and to build these data set huge data centers. And
00:07:26
Speaker
That's coming off the balance sheets of the biggest tech companies in the world and increasingly there's a lot of loan financing going in there. So that can, and and at the same time, that has driven up the value of certain companies. I mean, not the traditional enterprise SaaS companies, you know Salesforce and others are not an SAP, not where they they might otherwise be, but certainly the memory companies and the chips companies and so on. And some of these companies have gone up in value three, four, five, six times just over the last few years. So that could be, ah to use an English term, bubblicious.
00:07:57
Speaker
But on the other hand, that could also be the case case case completely rational, because if the demand for those products from real customers paying real money is outrageous, their backlogs going to grow, their prices are going to rise, investors are going to realise they're sitting on top of a bottleneck, and the share prices will go up.
00:08:17
Speaker
And so that all boils down to one thing, which is our customers, consumers or enterprises, spending money on AI Is that legitimate dollars they're spending?
00:08:30
Speaker
Are they doing it voluntarily? Are they going to continue to do that? And the thing that you have to really watch in this market, therefore, is what is the revenue that's pouring in at the top of the funnel?
00:08:41
Speaker
And I think we've seen what's happened over the last couple of years and what's happened this year with Anthropic. The revenues are pouring in and customers are spending more and more.
Reskilling Talent and Integrating AI into Enterprises
00:08:50
Speaker
ye Yeah, yeah. And any technology that's driven because of consumer adoption leaks into the enterprise, which is where I live, which is, and I call this the AI velocity gap, which is, as you're describing, the impact on our individual experiences in life are tremendous.
00:09:09
Speaker
How we write, how we think, how we organize, how we decorate our house, you know everything, right? But then you go to work on a Monday morning and you have to contend with archaic systems and SAP and things like that, right? um Closing that gap is, I think, the big next thing. and And it's interesting what's happening out here is you've got Dario and Altman. They're trying to sort of tone down the the the fear the fear rhetoric with AI because they've realized they are dependent on you know massive and global 2000 investment in their products. And enterprises are just woefully behind individuals right now in adopting AI and and the way that moves. so this catch up is really starting.
00:09:55
Speaker
and And I'm with you. I think we're at the beginning of a J curve, which is gonna dip a little and then it's gonna have as you said, exponential impact in the future. so so If you think about this, if you think about this and what what's behind you know the fact that these companies have set up um professional services arms, they call them forward deployed engineers, FDEs, they're basically systems integrators being paid much, much more than a traditional systems integrator.
00:10:29
Speaker
And what that's saying is that in order to really get a lot out of AI, you can't just slap a chatbot into the enterprise and hope that that changes things. That you really need to get into the gnarles of what what it is for that company to be that company. And and you need ah yeah smart people who understand the AI, but also can work with the customer to make that work in a way that is deeper than than how traditional systems integrated your traditional world has has operated.
00:11:05
Speaker
And I think you can ask the question, is that a bullish statement? In other words, is that a sign the market still has further to go? or is that a bearish statement? Is that a sign that this technology is isn't working. Now, for it to be a bearish statement, a lot of things need to be true.
00:11:25
Speaker
So then it needs to be the case that companies are finding it difficult to use AI and to to want to increase their their spending. ah And the cut the the the AI companies, the startups,
00:11:42
Speaker
are absolutely demanding and intent on selling to those companies who are finding it hard to use their products. And therefore they come up with this FDE approach. And normally a startup would say, well listen, if if the customer can't use my product, maybe they're the wrong customer, we'll go somewhere else. So a lot of quite odd things need
Current AI Landscape: Bullish or Bearish?
00:12:01
Speaker
to be be true. We know from Anthropic that enterprises who spent a dollar with Anthropic last year are on course to spend $5 with them this year. So this does not look like rational actors across the board, the thousands of big companies who are spending money with ah with Anthropic,
00:12:18
Speaker
ah doing that unless they thought they were getting something out of it. The bullish reading is that we've picked off the low-hanging fruit. What is a low-hanging fruit? It's meeting summarisation. It is pulling together a quick cash flow forecast. It is ah helping me draft an email. it is...
00:12:38
Speaker
Compare two documents. It is analysed for vendor proposals to figure out the strengths and weaknesses of each one. That's all the trivially easy stuff that, ah frankly, everybody should have been doing since the middle of 2023. All of that has been picked off and the management in those companies is starting to see the potential, which is why they are spending more and more money, five times more just on Anthropic.
00:13:04
Speaker
Then... In order to make that customer successful, the vendor will put smart people in. We used to call them pre-sales engineers or post-sales engineers when you and I were sort of in our 20s, right? That's what we would do. you'd You'd sell a piece of software, it would come on a CD-ROM, and to make sure the customer bought it and paid their maintenance, you sent a sales engineer with them to make sure they did it right, and they would be paying you for years and years. This is the same process. So the bullish signal requires you to believe a lot less than the bearish signal, right? To believe the bearish signal, lots of people have to be acting irrationally or pseudo fraudulently. To believe the bullish signal, it's like, oh yeah, they saw it and we're just gonna believe the numbers they tell us that everyone tells us and we've got precedent in our own very own industry that this is exactly what happens at these moments.
Challenges in AI Adoption and Cultural Shifts in Enterprises
00:13:54
Speaker
So I do think that the way I read this is um is that customers are finding it hard to get the gold medal. They've got the bronze and they've brought in a coach to help them get the gold medal because they think it's valuable to get the gold medal. And that seems fairly reasonable.
00:14:13
Speaker
Yeah, yeah. And so much of this is um people driven, um which, you know, the technology is already here and it's getting better. It's going to commoditize. The real issues with the enterprise is creating a culture where employees feel secure to experiment. um They are upskilling themselves and being trained well in figuring out how to resonate with different data points. How to say, look, I need to understand our sales forecast, align with different regions. I want to do some exploratory analysis of markets. It's starting to use the resources of your company to be smarter, slicker at how you operate. And... I just think we're on a massive talent reskill at the moment. And I love talking to the kids in college, lot of my nephews doing it, and Phil in Quantum at the moment.
00:15:06
Speaker
and And I'm just like, understanding how all the big tech firms are circling around the smart PhDs and students, because they're desperate for kids who just think differently.
00:15:19
Speaker
They're thinking about how do we solve or How do we look at this company's DNA and figure out how to do things more intently for it? So you talk about FTEs. It's like taking... um an engineer in the Ferrari garage and putting him in the car itself and saying go run a couple of laps and really experience what you're building.
00:15:39
Speaker
Do you know what I mean? It's doing and experiencing rather than just saying come up with a nice strategy on a PowerPoint. it It is that. I think the...
00:15:51
Speaker
The fact that there' the the the hiring standard bar is so high tells us quite a lot about the ah the complexity of what of what we're dealing with. i i think it's very hard for companies to really get to grips in any meaningful way with with AI. we We just published this essay um recently about why AI isn't showing on on the bottom line right right now in in more than yeah most companies. There's a really trivial answer, which I think is the sort of most historically accurate answer, which is it's a J curve as he's described. We should be in the bottom of the J if this is a great um technology. If it's a really easy to use technology like a photocopier, we're not much changed in the business. We should be out of the J. It's a really important technology. We're deep in the J. But the harder question is, how do you even navigate your way your way through that?
00:16:49
Speaker
And what we discovered within my my team, small small team, but it's I think it's simple similar to what else is happening, is that individual productivity creates superpowers in ah in in individuals and they produce more and more material and they create congestion.
00:17:11
Speaker
More PRs to review, more ah contracts for you know legal approval and those processes and the people in those processes need to be able to deal at that at that pace. And so you have to start to solve this this this problem as you go through a company, even if you get over cultural resistance, if people are just not used to working in this way, as soon as some teams or people in teams do, you will get this mismatch, this impedance mismatch. And we've seen it within our within our teams as well. in The company is small.
00:17:45
Speaker
and And then you get to the thing that you described, which was, oh, I would want to go off and do a market analysis. Well, the assumption in most companies is that the people who can do market analyses are few and far between.
00:18:00
Speaker
ah and And no one else has the tools and the resources to
Future of Enterprises in the AI Era
00:18:04
Speaker
go off and do it. Well, now, with ChatGPT subscription, you can go off and do that. But the company hasn't been organized in such a way that it can deal with...
00:18:16
Speaker
30% of the but the the employees doing market analysis and get expect to treat them sensibly or with any degree of respect. So these are some of the really, really hard challenges. And I'm not sure, by the way, that FDEs, forward deployed engineers, address that type of question because it goes into the heart of what a company is.
00:18:35
Speaker
Yeah, it's how can how can you scale ordinary and not just extraordinary, right, with people? Because, yeah, you can go out and hire entrepreneurs a crack group of PhDs or master's students, but then what about your regular regular people who are who are getting smarter and slicker using this stuff as well? So I think i think the FDE is a good starting point, but it really was something that I think was invented by, oddly enough, SAP in about 2010, where it was a model for how can we get more with smaller numbers of people to deploy software. So I think it's a good concept, but it's something that needs to be watered down, scaled by companies. And just this new way of working needs to become recognized.
00:19:25
Speaker
so so Can I just, i mean i' mean I'm going to challenge you now on something, Phil. I'm going to flip our, and so let's play this game. the talk You spend a lot of time with enterprises and I think that's why you sort of kind of cut through the noise so well.
00:19:41
Speaker
um if if if my if the if If an organization's people are going to get smarter and slicker, um what's goingnna what's going to happen? So you're in a you're in a market that's a competitive market, you've got three or four major competitors, you get smarter and so slicker, but you're paying for AI.
00:20:01
Speaker
So you've got higher costs and out of your smartness and slicknessness, you are now winning a little bit more business, which is great. So your your CFO is happy.
00:20:12
Speaker
Your three competitors, are now losing market share and their profitability has come down. And they look at you and they go, oh yeah, he he got Phil got Copilot or Claude Code. Let's get Copilot and Claude Code because anyone else can get there. And that advantage you have is going to get competed away. And the margin in expansion that you enjoyed is going to get competed away.
00:20:35
Speaker
And within a couple of years, you and your three competitors are full of smart and slick people. We're going be the smartest and the slickest of everyone. What's that film with Jim Carrey, Slicker and Slickerer? Dumb and Dumber. Right. Dumb and Dumber. Dumb and Dumber. Slicker and Slicker. Slick and smart.
00:20:52
Speaker
Everyone is slick and smart. So there that that that market share that you gained, you've you've given up. The margin pricing about power you've got you gave, ah you had, you've given up. But you're now paying...
00:21:04
Speaker
a fee for AI with within there. And so in fact, your costs have gone up and you're standing in the same place and you enjoyed a short two or three year temporary advantage, which if you're the CEO of this company, it's public, that is absolutely the moment to to leave when you're at a high. right but But so why isn't that what happens at this point?
00:21:23
Speaker
It's very good question. um And a lot lot of it's about competitive edge. What is it that is giving you the edge in the market? Okay, you got an edge because you got you've got to be able to do things quicker and sharper.
00:21:38
Speaker
So I bought a car insurance policy from a well-known brand a couple of years ago, and another well-known brand literally came along and offered me the same policy at half the price.
00:21:49
Speaker
And I actually knew the CIO of that company, and they've just been doing some really cool investments in AI to allow them just to run insurance operations a lot cheaper than the than the others. So, eventually, the other guy will probably catch up and they'll come back to me next year and say, look, Phil, we want you back as a customer. We can match their price. and Then it's down to, I think, things like experience, um how you work with clients. I think the human element is going to become more and more important.
00:22:17
Speaker
I really believe that. There's a very good article from Ethan Mollock, don't if you read it this week, about being human. It was good to coming from him. um But the ability to create client relationships, give them an experience, creates stickiness in your business, whether you're selling research or credit cards or mobile phones, whatever industry. And I think i think we get down to different ways of working with people who who want to deal with you in a certain way. I also think the nature of companies is changing dramatically.
00:22:48
Speaker
And we'll see a lot more small to medium businesses in the next two or three years springing up than ever. Because companies don't need to be as big. The Global 2000 is really struggling because most of these companies are just too big.
Job Creation and Workforce Preparation for AI
00:23:03
Speaker
There's too much legacy. They're struggling with change. And on top of that, if you're working for a great big bank or something, why should you change? mean, you're just picking up a salary and doing your job. and what you know So i do think
00:23:19
Speaker
if you're coming out of college in the next couple of years, you're going to get job offers from a lot of small companies who are looking to run marketing with two people, run sales with three people, run finance with two people and good software. So I think we're going to see a whole change in makeup of businesses, size of companies, ways of operating. um Is it going to create more jobs in time? Hopefully, it's difficult to say. um But jobs go where the demand is. Like we have a massive shortage of people in healthcare, care huge shortage of plumbers, electricians, all these sorts of things as well. So, you know, I'm not so glum about the future. i do worry about companies that are too big, right?
00:24:07
Speaker
I'm going to, this is mean that I've turned the tables this way, but I'm going to do it So what you smuggled in there, though, was you smuggled in with your insurance example, the fact that somebody could sell the insurance at at half the price, which means their cost base has to be fundamentally different. And in a business like insurance, a lot lot of that cost base is in headcounts as people. yeah and And so that sits uneasily with art where we started, which was enterprises retraining people. Because you you are yeah training is expensive, and if your cost base is coming down and you're going to bring down your number of people, it's not clear that you need to...
00:24:47
Speaker
you should you should be training those people. Like as a fiduciary duty of ah the board will say to you, wait, you trained all these people and you had no intention of maintaining their their their employment. That's inappropriate, right? Unless it's part of the you know the transition package that you're giving them. so So what's very complex here, I think, is that that There is going to be some kind of a reconfiguration of companies.
00:25:13
Speaker
And I don't even think it is about a question of of cost. and Cost where is going to be there. There is going to be a corollary correlative impact there between kind of cost savings and and and and people. But it's just that the nature of the work that you need to get done...
00:25:34
Speaker
is is going to change. you know In a world with ah where factories were powered by by steam, there was a guy whose job it was to fill the the coal boiler with coal.
00:25:47
Speaker
How many people in that job does HFS hire, employ, Phil? You can be honest here and you can admit that number to me. right um Zero.
00:25:59
Speaker
Zero. I hope. Zero. Right. And and but we don't we don't not employ coal people fillers because of the cost. We don't employ them because it's not appropriate to hire people in roles in that don't exist anymore. And I think one of the really hard things that we're going to have to contend with is that it becomes really, really clear that the reconfiguration of companies is going to take place because this technology moves information around in a different way. It allows decisions to be made at different points. And, and you know, as you rightly identify, Global 2000 has got very, very big. And this is the bit that I find I'm glad I'm not a public company CEO because
00:26:40
Speaker
How do I square all of these different signals? I mean, many of which you've said, yeah, global 2000 are too big. You've got to compete with innovation, dot dot, dot, right? I mean, this is a really, really knotty problem that they face.
Insights from China's AI Development
00:26:53
Speaker
Oh, it's it is. It is. And i had I had four CEOs on stage at our summit last week. I'm not going to name names. and who One of them said to me behind the scenes, if anyone is kidding themselves that we are going to shrink, they're kidding themselves, right?
00:27:11
Speaker
a lot of these big firms are going to get smaller. it's just It's just a changing nature of delivering services, such as IT services business. It's just a changing nature. You're just not going to need that many people. You don't need as many coders.
00:27:26
Speaker
You don't need as many people to answer customer calls. right um So big, some firms are going to share. They might share 20, 30% of their staff, maybe even more, some of them.
00:27:38
Speaker
But some of these firms will diversify into areas where there is new demand. where it could be supporting companies with their marketing or with their supply chain or with other areas. So the smart businesses are going to be gravitating towards areas where they can keep growing.
00:27:55
Speaker
And the new statistic we need to look at is how is growth tied to job creation. Right now, job creation on Wall Street is a dirty word. right it's You talk to a CEO of every services firm right now, they are so depressed because Every Wall Street analyst hates any company that needs to add people to make money. And when Jeff Bezos came out and said, I'm going to double avan Amazon's revenue in five years without hiring any more headcount, his stock went up. you know That's the goal of the company. It's how do we grow without adding people? Now, the next problem is going to be only 6% of Global 2000 companies have done anything remotely mature with AI. We've got the data to prove it. Nine out of 10 companies are just in pilot mode. they really It's a lot of lip service. They haven't figured this out yet.
00:28:48
Speaker
You've got this massive pressure point developing where these enterprises need to close the gap. And then you've got our lovely friends on the frontier models who are going to get frustrated because they want Global 2000 firms to keep spending more and more with them. So what's going to start to happen is this realization that to transform an enterprise from 2015 mindset to a 2027 mindset is going to require talent and people and skills.
00:29:20
Speaker
And there has to be an evolution of who's actually getting the talent, creating talent, developing the culture. there They're going to be the successful firms. Now, I think it's a responsibility of government as well to incentivize enterprises to train their people to incentivize academia to make sure their graduates are coming out with the right skills so they can be easily trained and developed and then i think big companies in services or consulting or these other white collar jobs they're going to take on graduates and maybe incubate them for two three years um don't build them out just have them learn have them develop and then
00:30:03
Speaker
you know I just think this whole rethink around talent is what's just starting right now. And it's going to be painful, but it's going to become political. and um But the bottom line is, is the world's moving in this direction. There is no stopping this at this point.
00:30:18
Speaker
And there has to be a smarter way of aligning where we're going with directions and decisions and policies. So on that note, you came back from China recently, Azim. Do you want to tell me a bit about where your experience was there and how you compare it to everywhere else?
00:30:34
Speaker
Yeah, I've been to China twice in the last ah nine months now, I reckon. and And this last trip was was fascinating. we were with a group of other writers and analysts specializing either in questions of of AI or in China or in some cases the intersection.
00:30:56
Speaker
And we had a over a week where we met people from 14 labs and robotics companies. So, you know, colleagues met with colleagues, Alibaba and... ah you Minimax, ByteDance, DeepSeek, Unitary, the ah whole caboodle, had long conversations, often dinners and lunches with them.
00:31:25
Speaker
One case, we we went out for karaoke and they left us at 2 a.m. on a Saturday morning to go back to the office to to check how our training run had been going. It was a really, really remarkable and very open conversation.
00:31:42
Speaker
And what I would reflect on is that yeah we I think we know the story that Chinese are short of compute across their AI industry because of various export controls. And they have really emphasized the open source approach for for development, which has been a key part, actually, of China's software industry for for decades.
00:32:05
Speaker
the But the thing that that I took away, and i again, I wrote this in in an essay, that that The industry within China is developing a really deep capacity for being lean.
00:32:20
Speaker
ah They don't have the compute, which is the the thing that you really, really need. And by they don't have the compute, they have about one seventh, one eighth of the compute the Americans have.
Global Influence of AI: Comparing China and the U.S.
00:32:30
Speaker
and ah the American labs have. So it's really, really, they're quite far, far behind. And you've got think about how that would otherwise slow you down because you can't essentially run a big training run and serve all your customers by inference all at the same time. You know you have to choose one or the other.
00:32:47
Speaker
And so they've established a discipline of developing models that are only six months behind the US models, um although they're two to three years behind on the amount of compute they've got. And i think that that is creating a moat that I call the efficiency moat, something that big is going to be persistently an advantage even as the compute availability opens up, which inevitably is with all shortages, it will over the next two or three years. And that, I think, is a really, really fascinating thing to look at.
00:33:17
Speaker
I thought back to the best parallel I could think of was in the automobile industry. So back in the 1950s and 60s, the Japanese moved from light machinery and motorbikes into cars, companies like Honda and Toyota,
00:33:33
Speaker
And they didn't have access to cheap energy or cheap cheap steel. They're still building up building their industrial base. Japan doesn't have many sort of indigenous energy sources. And they had very expensive access to capital. And in order to compete, they built cars through a process of persistently improving their efficiency. It got called Kaizen, it got called the Toyota production system, and it became a source of persistent advantage. So when in the 1980s, the US s started to customers, consumers wanted to buy more efficient cars,
00:34:10
Speaker
Detroit couldn't compete on a cost basis with with the Japanese. yeah Back then it was Honda, Datsun, and of course, Toyota. And that is a really important parallel that i to what I think is happening within China, which is right now, you know, the americans have got the American labs have got unlimited compute, although they actually don't have unlimited compute, they still feel very constrained. And they've got these kind of hyper-muscular models, you know, they're they're like um know a British bulldog that's been fed steroids like a massive barrel chest and they can do anything and but ah there'll come a point where enterprises won't need an AI model that's smarter than 40 Nobel laureates to determine whether a ticket needs to go to customer services or whether it needs to go to data protection right you can you'll be able to do that
00:35:05
Speaker
of triage with a really dumb, cheap, small model. So my sense is that the fact that the Chinese ecosystem has developed and is developing in a very different way with different attributes, even though there are commonalities, is actually going to be really beneficial for enterprises all around the world because it is avoiding us having this sort of you know monoculture where everything looks the same.
00:35:31
Speaker
Right. And how would you assess... based on your least couple of last visits, is there a different mindset approach to this that you're seeing elsewhere?
00:35:45
Speaker
I think the only places with really distinct approaches are the US and China and to some extent the UAE, which is a yeah much, much smaller business. But there there are things that are complicated in China and particularly think about the enterprise. So Chinese enterprises of all sorts of levels don't really like playing for software.
00:36:01
Speaker
um And they don't pay for software because it's ah it's like, you've you've written it for me, you're not doing any more work. Why should I be paying paying for it? And so they tend to work on project basis. So we we you know we know that there is a diffusion story in China, both within consumers, but also within within businesses.
00:36:18
Speaker
And I saw some absolutely superb, superb, superb examples of that. One was, I think you might appreciate this, is a brewery that was a self-optimizing, agentic brewery with this closed loop autonomous system being run by the kind of agents you and I might be playing around with. I mean, I don't know if it's necessarily open claw, but it was basically an agent harness strapped over ah an open source AI model. And this brewery had been running for a while, and it was very, very impressive. and and and it was But you know here are the things you have to think about, which is that in that industry,
00:36:53
Speaker
they're in that country rather they don't have a culture for paying for something like that on a retained basis on a monthly basis they will pay on a project basis than on a fixed but you know fix this bug bug type of basis so that that i think paints a picture of the kind of complexity and how different it is to the u.s market so getting back to know how we really look at this in in the enterprise, and I think this is a good comparison between China and the US, trusting AI agents to make real decisions, not just, you know, you said draft emails and things. I think that's one of the biggest issues holding back, you know big global corporations right now.
00:37:38
Speaker
What do you think needs to happen to engender more trust um from business leaders to put more confidence in these technologies? the The fact that they are so-called unreliable technologies which which you know hallucinate or confabulate, and we have all had experiences of those. The more you use ai the way I use it, hundreds of millions of tokens a day, I'm dealing with these things all the time and I figured out how to build verification loops and so on.
00:38:11
Speaker
um That, I think, is used and as an excuse for not really, really going for it. Just let's play a hypothetical experiment. Imagine you could have model as powerful as Claude Opus 4.7, which is seen as the most powerful model that was 100% reliable, that wasn't
00:38:34
Speaker
sycophantic, that didn't lose track of what it did was trying to do, that didn't try to over second guess you, but was just brilliant and 100% reliable. Do we really think that the CEOs of Global 2000s, who by and large have not spent tens or hundreds of hours themselves using this technology to really have the holy shit moment that you need to have,
00:39:03
Speaker
Do we really think that they would be moving faster? And I don't think they would be moving moving faster because all of the things that we have discussed so far, which are about the people and how do you retrain people and what happens to the shape of the company, what is the business that we are actually in,
00:39:22
Speaker
These are not to do with the quality of the models. At no point did we say, and this is happening because GPT 4.5 hallucinates. We never talked about it. It was about the technology at the state that it was in. And so so the barrier in the enterprise, especially at the leadership level, is is is nothing not to do with the maturity or not of the technology. It's absolutely 100% everything to do with the fact that If it's not a general-purpose technology, if it's like a photocopy or a coffee machine, the CEO doesn't need to know about it.
00:39:57
Speaker
Why would they? If it is a general-purpose technology, then these are epocally complicated things to deal with. And they take time for the norms and the standard operating procedures and the belief to emerge in in in an economy to to act on them.
00:40:14
Speaker
And the reason that I think it's the second and not the first is because I've happened to think for you know very reasoned reasons that AI is a general purpose technology. I also happen to think it doesn't really matter whether it it it isn't as good today as it will be in five years time, because frankly, the first cars made by Mr.
Adapting Enterprises to AI as a General-Purpose Technology
00:40:35
Speaker
Benz, the first engines were not as good as the engines we have today. And that didn't stop us buying them and using them 100 years ago. ah it is absolutely to the to down to have the senior execs
00:40:47
Speaker
contended with the fact that for the first time in 100 years, there's a general purpose technology as as disorienting as as this. And I don't really think they have because this is not something you can be taught about in a briefing. This is something you actually have to have to do And,
00:41:05
Speaker
you know you you for all the things that I do, you know, I make big kind of allocation decisions. I make important decisions that are important to me, but in you know in the external world, it will be seen as decisions of a certain scale. And I got my hands dirty because the oat because I knew it was important. And the only way you can actually understand it is if you've gone out and built these things and built with them and really, really started to to play about.
00:41:29
Speaker
That's the only way now. In two or three years' time, there will be more case studies coming out of business schools and there'll be you know more books of of experience. And then I think you will start to see companies leaning leaning into it. But I absolutely don't think it's about model quality right now. And I think it's absolutely to do with the fact that the Global 2000 companies have enough momentum, that there's enough inertia that faced and an uphill struggle that faces any real disruptor to them, that CEO tenure is not sufficiently long to for for them to care. I think if you went off and said to the CEOs the Global 2000 companies, we are putting all of your compensation in a clawback,
00:42:11
Speaker
on the basis that if this company is disrupted by ai in the next 10 years, we're going to come after you and get it all back. I promise you they'll be moving quicker than they are now. Yeah, yeah. I think that's a very good suggestion. i mean, you think about it, you know employment traditionally was workers who usually maybe needed to retool once in their career.
00:42:35
Speaker
Now we're having to retool two, three times. So it's this gap between... week. how fast the technology moves and how slowly institutions can respond. And, you know, you're exactly right. is If you are very entrenched in the technologies yourselves, you just know when you talk to people, whether they are or not, or whether it feels like a conversation we had 10 years ago where you could tell,
00:43:02
Speaker
I've seen exactly bullshitting. You know what i mean? way you could You could bullshit your way around a lot of this stuff, but I think we're at the point where we did our summit last week in the US, and it's a great litmus test for where we are.
00:43:15
Speaker
You could just feel the fear in the room for two days. Oh, could you? You could really feel it. like it's ah It's an anxiety. It's ah it's a... it's It's ah almost, you know, some you can tell some people have really embraced it and they're very excitable about it. And they're smart as well. they're Everyone's paranoid because that you can think, how useful am I going to be in two or three years' time?
00:43:38
Speaker
And then you've got people over 50 who might be just thinking, look, all i have to do is ride yourself for a couple more years and I'll just retire or something. and I know I look very young, Phil, but I'm i'm well into my fifty s And, and ah you know...
00:43:52
Speaker
There is a lot going on, right? um I mean, there is so much news. It is exhausting. There is so many changes to models and harnesses and this and that. It is very, very tiring, even if it's, you know, you've made it your job, so I have to look out look into it. and And there are moments where you just think, God, I wish, I wish this would would all slow down. um i think what you have to do is you have to adjust your expectations about this. And we all have responsibilities to...
00:44:21
Speaker
um yeah know even beyond all our work, right, to family, to children, to to to be given the beneficial positions that we hold in the world at large to get a sense of really, really understanding what is what is going on and what's possible um and and why and and how.
00:44:44
Speaker
I think it's you know it was always going to be difficult for big companies, for us to make sense of a general purpose technology and because it's happening to us. We tend to sort of single AI out for particular qualities that it that it has. but But I really don't think it's it's about that. i I really do think it is about the the fact that
00:45:15
Speaker
You know, a simple question would be, if everybody is more productive than no one's productive, so we've run to standstill.
Enhancing Productivity and Corporate Culture Amidst AI Changes
00:45:25
Speaker
And...
00:45:26
Speaker
And is that really the end state of using AI for a company, which is to increase productivity? I wrote this essay about this, about the final stage of um of what ah and what an AI company looks like, and it's the hardest one to reach, and we don't really find many companies, if any, who have got there, which is actually about this one metric.
00:45:51
Speaker
So a company has a cycle time, they actually have multiple cycle times, which is, ah the cycle time is essentially how long does it take them to observe a change in the external environment, in the market, in their customers, in the technology landscape? how How long does it then take them to orient themselves towards that change, make a decision about it, and act on it, and then go through that loop again? And you can you know companies in your ah arena, cloud starts to show up in 2005, 2006,
00:46:22
Speaker
Ignoring the legacy issues, it took some companies a decade to notice it and decide to act on it. So their cycle time is like 25 years. And my intuition is that the cycle time in the world of AI is going to be the thing that AI native companies are going to be exceptional at. And fundamentally not having that static, having it dynamic. In other words, getting it shorter and shorter and shorter and shorter. And that's how you construct persistent advantage. In the previous example where we talked about everyone being slicker and slickerer,
00:46:54
Speaker
your advantage gets competed away because your your competitor gets it and now you're all competing same margin higher costs because you're all paying for AI the way in which you turn that into something that you can compete with is the tightening of your response and learning loop as a company the cycle time of your company And if you get good at that, and AI will help you do that, you will constantly be able to but as basically be, as fighter pilots say, inside the and energy envelope or envelope of your ah of your opponent and therefore win the market.
00:47:27
Speaker
Absolutely. And honestly, the whole makeup of work is changing. um And so I run an analyst company. And my biggest issue right now is a lot of my analysts have got really, really good because they're all using these tools.
00:47:42
Speaker
And I need to make sure they're collaborating with each other more than they are with their LLMs. And it's sort of it's an issue. And me, myself, personally, I cut back on the amount of meetings, I tell because I want more time to work. I'm i'm i'm like an analyst on steroids these days. it's It's a lot of fun. I'm way more productive. I can get a lot more done.
00:48:05
Speaker
I can expand what I cover. um And the only thing I said to my wife this morning, I said, I'm bit worried I'm getting ADD. She said, me too. But it's true. i think it's like there's so much going on. I don't know about you, but the the workplace has become so manic, so chaotic. There's so much stuff flying at you. There's so many tools open for communication and stuff that learning how to manage and cope with these things is becoming more and more important.
00:48:35
Speaker
And I find I'm spending more and more time trying to just get good time with my clients and my staff because that human interaction is now more valuable than ever. It's like people are lonely, people are stressed, people are having more, hate to say it, mental health.
00:48:52
Speaker
issues because the way the nature of work become very intense. That as a leader and a coach, you've you've got to change your own style and adapt as well. So, you know, this is ah like a voyage of change.
00:49:05
Speaker
I love what you're saying about this time to retour, but i think this is much more personal than any technology revolution we've been through, probably since the industrial revolution. So,
00:49:17
Speaker
Well, yeah, I mean, I think that that idea that it's important and it's uncomfortable and you need time with your people is ah is a really powerful one. and we We're a very small, a small team and.
00:49:36
Speaker
But we have changed, the way people work has changed over the last two or three years. i mean, they do have more time, they have more time away from their computers, more time working by hand. And, you know, why why is that? It's because you can do high quality work and the AIs can take that high quality work and turn it into the the document or the pseudo finished product for you backcom to come back and and and polish.
00:50:04
Speaker
I spend much, much more time now writing longhand with a pen than I have since 1989.
00:50:16
Speaker
nine So I think the Berlin Wall was still up when I last used a pen as much as much as this and maybe, yeah, I don't know what kind of what's happening in pop music back then. and so that and That has been one of the ways that i i have been able to kind of get more out of the the use of of AI tools. and And we, of course, have got all of this these things running. we've We've built our own reasoning architecture, which by which I mean yeah lots of knowledge comes into the the company in different ways that we, are through our data products and research, we we we read and write.
00:50:57
Speaker
and we then have internal tooling so that if you want to reason through a particular class of question, you can just you can just fire that off. So for example, if you want to see how well a group of customers might respond to a new piece of research, you can take that piece of research, you can spin up a simulation of 50 customer approach personas based on kind of real customer interactions
00:51:27
Speaker
And you can watch them argue over the piece of research or these agents that represent them and and go back and look at what they clung to or didn't cling to. Decide yourself as a human what's important for me to change or what is working really, really well. So we have all of these types of of tools available for the for the team and there are there are sort of other things in there as well. And...
00:51:51
Speaker
and that's all That's all good and well. And I think you could probably run a business like ours without anyone, frankly. i mean we could all go down to the beach. it Except that that our business isn't a ninth ninth decile business.
00:52:13
Speaker
We're a tenth decile business. and And so being the top 1% of what we do, feeling that we've been at the top 1% isn't sufficient, let alone top 10%.
00:52:24
Speaker
And that requires us. And that requires us. And so we you know we will never eliminate the bottleneck problem because... because the fact that I'm typing the words in by hand or the fact that I hand wrote them with a fountain pen and dictated them into, you know, voice to text is a really, really important part of our process of our product. And most importantly of our promise. That's right.
00:52:51
Speaker
Having an authentic voice has never more important in my my view.
Conclusion and Farewell
00:52:55
Speaker
Well, in that regard, um, I'd like to thank you for a wonderful hour. I really enjoyed this conversation.
00:53:01
Speaker
Um, We never had to stick to the script. And it was great to hear how you're doing. And I hope to see you soon. We're doing our London Summit coming up in a few months. So I'm going to see if we can get you along for that and you can see how our world has changed a bit.
00:53:19
Speaker
Yeah, that's fantastic, Phil. It's great to talk to you. Thank you. Terrific. Love the time. And to the next time. And I look forward to airing this this podcast with everyone. It's going to get some good feedback on it.