Daniel Dines on Why AI Can't Replace Humans in Enterprise & Why Europe is Falling Behind

In my opinion, even the labs in China that are building AI right now, I would classify them as the good guys.

I've never hidden from my employees that there will be a transformation. Jensen is bound by the success of open source.

Models are interchangeable, but the workflow, the map of work and the workflows around the map of work is where the real value is. Ready to go.

Likewise, dude. It's always a pleasure to be here and it's the I think the hottest moment in technology. So I'm very excited to talk to you and get your perspective also on a lot of topics.

Man, I dreamt to write a book since I was a kid and I discovered I have no talent and now I got really a great opportunity. Claude and ChatGPT helped me a lot. They were my ghost writers and it was a good moment actually to put my own ideas in order because really when you write something you get much more clear perspectives on what you are doing. So it was a maybe almost 6 months effort doing this and I started with different threads of thought.

One was, you know, what are the limitations of AI? Is there any durable limitation of AI where in a couple of years there will be millions of Einsteins in the data center and we can all go to play or do whatever we would like to do because the Einsteins will do the work for us.

Look, when I heard about this statement that in a couple of years we will have millions of Einsteins in a data center, look, I was really concerned. And Dario is a guy that I highly respect and he's highly successful in what he did. So I was thinking, what does it mean for us? What does it mean for, can I hire one of these, put it in a laptop somehow, assign, you know, an enterprise account, a Slack account and everything and ask it to do my job or do whatever job in the enterprise? It seems that reality is maybe different and probably Dario wanted to say that we'll have millions of entities that will have some of the reasoning powers of Einstein, which I agree, but not Einsteins as persons, not Einsteins that are capable of learning on the job. Because do you agree that one of the major expectations when you hire someone is that they will learn on the job? There is no manual that a company has to give a new employee, this is exactly how you do your job from end to end. So we expect that.

But you say AI learns on the job. AI can create a notepad on the job, a scratch pad where they can memorize some of the policies on the job, but AI doesn't alter its weights on the job in the way humans are transformed by a job. This is a huge difference because let me give you an example. You have two chefs. One chef 20 years has done only Japanese food, the other only Italian food and you give them one recipe. Okay, they will create different food. So it's not that you can write down your enterprise on a sheet of paper. It's much more complex. It's a becoming. Think about if I give to someone the ability to read all the books about chess, do you think he will become a grandmaster without playing, without losing, without going through all of these processes? Probably not. If I have you watch all the videos about skiing, are you becoming a skier? No, you're not becoming a skier.

I disagree with you. I think most of the people will display some sort of micro-initiatives during the job. Maybe I have a hunch this customer is going to churn and I can act before even any data is coming. How do I develop this hunch? It's through my years of transformation. It's not written on a piece of paper. It's a big difference between writing an operating model on a piece of paper and living it. It's almost impossible for an enterprise for AI. Do you admit that everything has to be written down, documented? Every time I'm asking a question to the model, the model will read my entire enterprise, will have to read it. Okay.

So this is not possible.

This is one of the biggest bottlenecks right now because AI doesn't train on the job, doesn't train their own weights. Every time I am doing something after this talk with you, I am being transformed and I carry with me this discussion in all my thoughts. This is not true about AI.

It has memory, but memory is not necessarily learning. It's not the same thing. It's memory. It's just a thing that is written down. When I'm talking to you, I don't go back into my memory. I am just being transformed. It's like a model is like a new version of the model that comes improved. That new version is not the old version plus a piece of paper that has memorized. It's transformed into its own weights. This is why a model becomes much better. Look, I want to give you a simple example on using our own technology. We make our UiPath platform available for coding agents. So people are making it's much easier to create automations on UiPath right now. But what we discover is that a model that has read open-source technology and they have a lot of examples and they already have in its weights, you know, a certain technology will be much better than to create on our own technology because regardless how many prompts and skills we create, the model has it in its own weights. It's very different. And think of all the metaphors in the world when you read something versus when you live something. You can read the biography, it doesn't mean you live that life, it doesn't mean you are transformed and you are going to answer like that person that lived it. So I think to me really this is the biggest limitation that the models have right now.

I don't know, then maybe we are the result of a self-improvement loop because look, do this thought exercise. Let's put the model that we have today with the best technology, put it in a spaceship and throw it to the stars. Let's say that we have this technology like I think von Neumann imagined that is self-replicating. This spaceship goes to different stars, gets energy and they can continue. So compute will be infinite and models will self-improve. Where they would end up with, maybe they will create a simulation of a world like ourselves, right? Because they will intrude infinitely, basically this is the theory. So they will simulate a world as complex as our own world within it. But that means that we are a part of infinite simulations. So look, the answer is, and I don't know where it's going to lead, but I know that there is a big distinction that I made in the book between will and reasoning. And it's not like that we are certain that the will to do something, you know, emerges from reasoning or even from consciousness. I think will is a kind of a separate, you know, part of the fabric of the universe that I don't think we as humanity have a clarity about what will is. I think it's wishful thinking to believe that I can take a big model, put it into a self-improvement loop and this model is going to generate will. I don't kind of believe so.

I would say if they truly believe that this technology is becoming rogue and they cannot control it and their experiments will create significant loss for, I don't know, internet, other systems. If I were them, I would pace it at any cost because I don't want to risk going to jail. Honestly, I think there are laws that kind of control this type of rogue behavior. So honestly, I don't need an external pressure to control. I would be just a concerned citizen and I would not build a technology that is causing harm. But now of course they think that this is the only way to protect against the bad boys. So we are the good guys. But there will be some bad boys that, you know, probably in other parts of the world that will build the technology regardless. So I think that probably they ask more like a pass. I want to build this technology at any risk and I'm willing to open, you know, my gates for others to see how I'm doing because I want to do it in as good a manner as possible. But at the same time I want to be free of consequences to me. I think this is a bit how I read this memo. And because otherwise, guys, guys, I think it's kind of obvious. I don't think we can, we will reason with the bad guys and we can make a coalition with the bad guys to stop the frontier. So we can make a coalition only with the good guys regardless. I would take even, in my opinion, even the labs in China that are building AI right now, I would classify them as the good guys. But someone to me, I think the indirect attack is probably on open source because basically they say even if the good guys are building open source, but that open source will get into the hands of the bad guys and this is unknown bad guys, this is the real danger. So the danger is in open source. So indirectly it's also an attack to open source in this way. It's a way of interpreting, I guess.

I think so. Yes. I don't think people are scared that OpenAI will build a competitor to them necessarily. I don't see this coming. I think they are more scared that their IP would leak to other existing competitors. Somehow, because I don't know if I'm manufacturing somehow shrews or whatever, I don't think OpenAI is going to come and compete with me on this thing, but probably some of the other guys can get indirectly the other models the same intelligence if I will train the model. I think that's the real danger and I think it's a legitimate danger. And I think everyone is trying to protect their IPs.

It has become very clear to me that another limitation of AI is what I call exactness. And AI, by its nature being probabilistic at every step, in a way it can lose it. You know, while you do 100, 200 steps by AI, even at each step you have 99% probability, for instance, it's, you know, 0.99 to the power of 100, you will end up with maybe 60% probability to do the entire step right. So AI doesn't have this mechanism to follow steps exactly hundreds, millions of times in the same way. You can see it even if you ask AI to multiply very large numbers millions of times, at some point they will make a mistake. And also it surfaces kind of another simple idea: even if you have a tool like AI that is capable of doing multiplications, why are you not using a computer that is doing these multiplications millions of times, you know, 100% of the time exact? The fact that the tool can do a job doesn't mean you have to use that tool to do that type of job.

I think that ChatGPT acts like an interface to convert my questions in natural language into exactness. But the exactness is not run by ChatGPT. Exactness is run by a computer. Because even today if you ask ChatGPT, please multiply these two big numbers, they are using behind the scene a computer and they will give you the exact number. This is part of the power of the models. So you actually can see on the desktop level on this type of code work or judge work, you see the capability to call tools. And what I am saying, you extend this capability to the enterprise level when everything that should be exact should run on exact technologies. There is no point to run it on probabilistic technologies. And here comes, I think, the most interesting thing. There is an asymmetry in the deployment of AI and automation in an enterprise. Deploying AI agents is not getting easier today than it was two years ago, in my opinion. But deploying automation has become much easier because I can create these automations with AI with coding agents. Coding agents have been the most major giant leap that we were seeing in the past year. I would say since the invention of ChatGPT, then chain of thoughts, and then coding agents were, you know, the major milestones. So with coding agents that act in kind of a design time, when I build the systems, I can create automations that work with exactness every time during the execution time. So this is the asymmetry that is happening right now. Plus, when an automation breaks because any change in the upstream system, AI comes again into play and fixes the automation itself. So this is really the pattern that we are seeing emerging in an enterprise, and this is that AI is actually creating the software that runs an enterprise. And this software really cannot behave rogue because this software cannot change its behavior in real time. It's not a probabilistic technology. Even if the software is created by AI, we can really audit. We can have humans that read it, validate it. I can have tests that, you know, for a certain input will guarantee that the software will behave in the same way. So that again makes this pattern extremely powerful. You use AI to create software that runs the enterprise in a predictable, governable, auditable way.

Maybe more than a thousand.

Yeah.

Around 4,000.

Yes. Why?

It's a complicated question because I think I would answer more with the thing that is in my book. I think the more we transform our companies using AI, we have to in the same time transform our workforce. I've never hidden from my employees that there will be a transformation in the company. But I told them up front, guys, we are not doing anything stupid. We are not just using AI as a pretext to cut a part of the company. We need to do the transformation in the same time as we successfully adopt AI in an enterprise. And that's actually another point that I discovered, you know, writing this book about, if you look, I was looking deeply at jobs and what jobs can be affected by AI, what jobs can be enhanced and how this transformation is going to look like. Man, one of the things that seems very simple in retrospect is that people don't have very simple jobs that again can be defined on a sheet of paper. I think every job has some kind of an outcome that is measurable and this is the outcome that people are hired for. But there is another outcome that is part of the institutional strength. Think about my deep relationship with the customer. It's not necessarily part of the numbers that I'm producing, but it's maybe what makes this customer stick to my technology. This is a different outcome. If I'm going blindly and I cut, you know, on a number of A's and say this is because AI is going to replace them. AI can call the customers and write emails. I don't think AI can supplement the human connections and the trust. So this is a different outcome of the job and it reflects on every employee, on every type of role in a company. So to me, an enterprise should have a ledger where they actually understand what people are doing besides, you know, their main definition of the role. And only after they have this ledger and understanding, they can look at how does the AI transformation look like? What kind of jobs will be affected? How can I move people maybe from one job to the other because they still carry some kind of, you know, the cultural aspect of the enterprise. So I don't think it's a simple problem as AI is going to cut 20% of the company. Let's do a riff 20% out and then we increase the AI adoption. I think on the contrary, when you do this blindly, you risk to hollow the enterprise of exactly the same talent that thrived during AI. Because let me give you an example in the book, I call this like the credentialed middle that kind of was the type of people that were prevalent in any enterprises. And if you think of our education system and the way we hire people, we hire based on deep expertise in a particular domain, a credential expertise. And this is exactly the type of expertise that might not be needed so much that AI can really help. So you will need fewer of these experts, but you will need more people that have initiative, that are capable of maintaining a relationship with the customer, that can be mentors for new employees, that bear, you know, the cultural aspect of the enterprise. So it's counterintuitive because you will tend to cut those people that are not the biggest experts in the domain, but you will cut exactly what you will need to bring the AI to supplement these experts. So one thing that I...

I 100% agree. We will need less of the people in probably most of the roles. But the main question is which ones? How do you choose? I think you can choose quite simply for where is their verifiability, which is, you know, finance and accounting, it's quite clear what is right and what is wrong.

But the frame is clear. Finance and accounting, did they do your expenses?

This is one of the domains where the frame is not clear because when I receive an invoice or I receive an order from a customer, I can treat it differently. There are not always the rules there. I know for this customer, NVIDIA is going to ship with priority to OpenAI. Maybe they have a rule that says there, yes, my first chips go there. Maybe they don't. That rule they don't. If it's not captured in a frame, I think AI cannot learn it. This is why you need to create this manual. We call this manual the map of work. You need to hand the map of work to AI in order to be successful.

To basically capture how the work works, how the work happens in an enterprise. This is the map of work. It's all the workflows, all the exceptions, all the procedures, all the systems that you use in order to fulfill a goal of a process.

But you come to my point. Even in finance, you cannot replace everybody. So you...

Yeah. Look, you only need a certain number of people. The thing is if you have X number of people, how do you understand which of them stay and which of them has to go to maybe do different jobs? How do you know? Because ideally you will get the people that will have literacy in AI, that can display initiative, because when AI cannot exhibit initiative in the human sense. So out of this number of people that you want to keep, you want to keep the people that display the most initiative. Even a finance person treating an invoice with the customer contributes to the culture and how my enterprise is regarded. How can I make a distinction between that person and the other person that care less about how they treat the customers? Because I want to keep the first because the second is more prone to being displaced by AI. This is the ledger that I think enterprises have to create and to understand different outputs of people. They need to judge people by this kind of hard-to-define output.

I think this is the crux of the problem.

Yeah. Yeah. And this is where we put a lot of effort as a company and we are introducing a new technology that we call it like ctography. And ctography is a discipline. So it's a discipline to help companies surface all the information of how the work is done and help them create this map of work. And one big important of ctography is to actually investigate what people are doing on their desktops. And we have a product that we call the ctographer agent that can interview people, real subject matter experts, have them basically record what they are doing and the agent is interviewing them in real time. Like if you interview a finance person, it can ask why did you change this invoice when the zip code was different? Why did you choose a different path? Tell me and they can start surfacing all of these exceptions. So that's real agents interviewing real people. It's pretty cool stuff. And then you consolidate data from multiple people and then we create the process maps. So we show them how the work works in kind of real time. And then after this map of work that show the work as is, you can use our coding agents and you can come up with an idea of how you should transform the process. And you transform the process by printing software. So you start with a process that at point A it's kind of fully manual. Of course you have enterprise systems like system of records and everything but people operate the system. I think the goal of any enterprise is to have less people operating the system and more automations and agentic AI operating the systems.

This is inevitable. And so I think it very much depends how you pitch the company. So in UiPath, I think it was important to tell people again, we are not doing anything stupid. We are not doing any mass extinction, you know, for the precedence of AI, but transformation is inevitable and you guys have to transform and everybody will get a chance and the people that will become more literate in respect to AI will have a better chance not only here but in the future in any other job. I think that's the message that I think everybody should get. So...

I think the response is good. AI adoption requires more cycles than just discovering the process and people's input in this. Look, in the beginning of the year, I think the fear of people across the industry and not only in my company but in many companies was off the charts, the fear of completely being replaced. Now I think people start to get a bit more understanding of the durability of their jobs and this AI diffusion in enterprise that can happen at a slower pace and can happen one process at a time because these millions of Einsteins are not hireable yet.

Look, I personally don't care about it. And let me tell you something. It's a simple hypothesis. If the work at the quality of a human can be done by a machine, I will hire today a machine even if it's more expensive than a human because human cost will only increase and they bring errors into the picture while the cost of machine would increase. So I will have a competitive advantage compared to people that will stick to humans if that's... and I think everyone will do this. So I don't think the cost of tokens will be the real question if you replace a person with AI, but the real problem today is that AI cannot replace a person because if again, if bring me an Einstein that replace me and I will happily go on any vacation in the world, but this Einstein doesn't yet exist and probably I would like to be interviewed and have this podcast with another Einstein. Man, this thing doesn't exist today. That's the reality. So let's call it a reality. Let's call a spade a spade. Maybe this technology will emerge and somehow Einstein that embody like a person that have will, that gets transformed on the job, learn on the job, have the capability of reasoning, imagination of Einsteins exist. Of course all the jobs will go extinct.

I want to see this, man. I think I heard, you know, companies that replaced hundreds of support people in the past and now they are rehiring these people. I think until this model is proven and is proven at scale, not in a particular industry for a particular guy, I don't think we can extrapolate for one data point that is going to go across industry.

Did you vibe code tools out?

Look, we were trying to replace like a procurement tool writing ourselves and I think we have a lot of success and initially it was written only by AI, but then it comes to, you know, speaking to this self-improvement loops kind of don't trust. We don't have enough test, we don't have enough trust to put this tool in production 100%. And humans in our experience, humans have to intervene a lot into how this vibe coded tool, for instance, the database schema that vibe coded tool created was completely bogus. So a human has to come and, you know, create the structure. Right now you are not at the point where, you know, you will have like a business user that understands a problem and they will vibe code a tool. So you will still need to put engineers and people, you need to maintain it. So it's a long process. So eventually you will end up paying probably as much if not more as the tool you replace while you keep some of your good and best people bandwidth occupied.

I would buy Salesforce as a system of record.

As a stock. Look, I invest in software as a category and I think I made a good investment a few months ago because I bought in the bottom of the SaaS apocalypse. Even our own stock, you know, has been doing better. And but the markets today are driven so much about sentiment and not the value. So it's kind of hard for me to make a judgment of an...

individual company. But I don't think Salesforce can be replaced by vibe coding, if this is the question. I don't understand why a company would go public today. If you think about the two drivers of being public: number one is liquidity for shareholders, employees and shareholders. Well, Stripe and many companies are able to have liquid stock in private markets. And two is the ability to have M&A, a tradable asset that you can buy with. I mean, many, many private companies are able to buy with private stock. Stripe, we're going to do PayPal with private stock for 60 billion. So that's not a barrier. And the casinoization of private markets as public markets, as you said, with current stock markets being sentiment driven, I don't understand why one would. But Harry, let's not make a confusion between, you know, some very exceptional companies and the most companies that are out there. There are so many companies that are kind of zombies right now. These 2021 zombies, they would fare better in the public market right now. At least their investors will have a way for exiting. Their employees will have a way to make some money. Nowadays all of them are sitting on paper money.

Okay. Public markets will face them with the reality of, you know, what's their real valuation. So I would not say there is no value in why Anthropic is trying to do an IPO in the end.

No. So maybe I think some will do, but some will do, of course. But I think we will see an exodus of... And I honestly, I don't believe at two trillion or whatever, maybe they would reach five trillion, because if I buy at two trillion, I need to have a path to five.

Anthropic in particular, I wouldn't. Because do you remember on our last podcast, I think you asked me what's the company that I bet on, and I said Anthropic. And Anthropic was worth 60 billion market cap. Maybe I was stupid. I didn't invest.

Yeah. Yeah. 100%. So, look, I would not sell. I think it's the same with OpenAI, man. I think OpenAI has caught up quite nicely, and I use them interchangeably right now. I just think we're in a market where the big get bigger and value concentrates more than ever. Actually, more than ever. But the real question is, Harry, would I buy at two trillion? That's my real question. And look, right now, I need to see their real numbers, you know, to understand if I will put money in their IPO.

But I think Jensen is bound by the success of open source, because if Anthropic and OpenAI... this is becoming a duopoly, and basically their TAM is in like trillions. It's basically the work.

Of course. I mean, OpenAI being Halapo and Anthropic are doing there.

So I don't think Jensen will be doing so well if they are the single biggest providers and the source of truth and light of God will come from only Anthropic and OpenAI. So therefore the open source should succeed.

And I think it was a great move buying Hugging Face.

Because I think it encourages... it hosts all the open source models. It's putting the money where the money is for his company.

I think Neatron is still a small cog in the picture. I think it's valuable, but I would not call it... it's not the same chip size as the others.

It can be, because every major infrastructure in history has been overbuilt. And I think it's a simple explanation, because I was thinking why every infrastructure is overbuilt. Because you have to make sure you get, you know, the biggest piece of the opportunity. If the opportunity is big, it doesn't matter, you build a little bit more than is necessary. So it's clear that now everybody... there is only 100% of the pie and people are building right now 200% of the pie. There will be losers.

Yes, but are we underbuilt to the extent of this trillions that comes into the infrastructure? I don't know to answer this. And everything happens on the premise that AI is going to replace human work in a really large scale. We need to see the timing of this replacement and transformation. It's a big difference if it's coming in 10 years versus the next two years. So I don't think it's overbuilt for the next decade, but it might be overbuilt for the next three years. And stock markets and capital, you know, can be merciless.

Yeah. Harry, you should read my book, my friend, because it answers exactly the same questions. So when the frame that a person operates in is really well defined by someone else, like in law, AI can be devastating. And the impact when the frame is not as clear and there are so many exceptions that are custom made for an enterprise.