AI investors are navigating ‘peak ambiguity’, says General Catalyst chief

Since stepping up in 2021 to run General Catalyst, one of the US’s largest and most active venture capital firms, chief executive Hemant Taneja has taken a wide-ranging and unusual approach to investment in artificial intelligence.

The firm has backed some of the biggest start-ups in the sector, including Anthropic and Mistral. But it has also launched an “AI roll-up” strategy to buy mundane service businesses and inject them with AI, and acquired a hospital system last year that Taneja says can be revamped with technology.

General Catalyst’s backers have rewarded the voyage into uncharted territory, with the firm raising an $8bn fund last year.

In this conversation with the Financial Times’ venture capital correspondent George Hammond, Taneja describes how the firm is investing that pot amid an AI investment bubble, and the risks if AI is not developed responsibly.

George Hammond: Let’s start with General Catalyst’s approach. You describe the firm as “an investment and transformation company” and I wanted to break down your approach to AI in both those regards. How does a firm like yours make money from AI? 

Hemant Taneja: Let’s start with the idea that AI is a transformative force and the value really accrues to companies that have existing customers and data, because they can actually be beneficiaries of applied AI. Organisations that have the culture and courage to transform themselves will be strengthened by AI; organisations that don’t will be left behind. It starts with that theory of change. 

Then let’s break down the transformation of a business into various components: one, does an enterprise have the infrastructure that can adopt AI? Second, do the language models work well enough for the industry at large or do you need to build small models on top that are required to address the needs of that particular business function?

Third is the workforce transformation: are you going to have an organisation that can be highly productive, with human beings as well as AI agents working in concert with each other?

GH: When you apply that AI framework, what are some of the opportunities that come out in a field such as healthcare, where you have done a lot of work? 

HT: Let’s take an operational example: you have a call centre, can you “AI-enable” that? One of the companies we’re building, Crescendo, is working on that. The other example is Hippocratic: you train an AI nurse and make it be exceptionally smart, safe, empathetic and very affordable compared with what it costs to have a nurse on the phone. What that does is alleviate the workforce and, if it’s cheap enough you can have a mindset of abundance rather than triage. 

Think about taking AI and saying “can the call centre be more cost-effective so the company or the health system can have operational leverage and be more profitable”. And can we rethink the value proposition to the consumer or the patient because our AI nurse is so cost-effective that we can afford to call everybody, with a much greater frequency and for a much longer period to take care of them?

If you generalise what I just said, by applying AI to a customer experience, you can enable every business to think with that kind of depth and how to take it to the consumer. That’s a transformative value proposition for the businesses that end up applying it.

GH: Are we talking about a value proposition or reality at the moment? There’s huge promise to applied AI, but there’s also a lot of friction from patients using virtual agents, issues for existing employees, and quality and fidelity questions around AI tools as they are now. Is this gaining traction in a meaningful way? 

HT: There’s a nuanced answer to that. If you look at the traction that Hippocratic is having, they are making hundreds of thousands, if not millions of calls a month, that nurses would have been making. People are being taken care of and the customer support costs of a health centre are going down. That is real because we built it in collaboration with the health system. 

If I take a normal call centre, if you look at the customer support companies that are going in and saying, “hey, let’s apply AI”, you will see mixed results. You have to update the infrastructure, add the models, transform the workforce, for AI to really take hold. That’s a very hard problem. 

Our theory of change was: the places where organisations have already decided to outsource, because of labour arbitrage, were the places to start. They were already comfortable saying “this call centre in the Philippines will make calls for us”. To outsource from that to an AI agent making calls is a much lighter lift.

That’s where the bright spots and the challenges are. There’s inertia. But again, those are great insertion points in the enterprise and that’s why these AI roll-ups are an easy entry point. 

GH: Could you capture in a nutshell what the AI roll-up strategy is?

HT: Take any department in the enterprise, what we have done is partnered with a strong technology team and acquired service businesses in those departments. That includes customer experience, legal, accounting, finance in industries like insurance, healthcare and others.

We’re going to find those companies that were traditionally service and labour companies and transform them with AI.

GH: Going back to your point about offshoring to the Philippines and then that being a natural step to AI agents, what are the consequences of that? 

HT: The places where we’re seeing this happen really fast, the consequence is that we’re essentially taking away the jobs in these emerging economies that made up their middle market. And we’re transforming those jobs into AI productivity opportunities for these companies that we’re building. 

We have to think about applying reskilling frameworks to what those people are going to be doing with urgency. That is something we’re giving a lot of thought to because unlike the previous technology trends, AI is the zeitgeist all over the world: every CEO in every industry, in every country, is thinking about applying AI at the same time.

That has never happened with any technology trend before. And we’re seeing the early evidence of that in these companies. 

GH: One of the things you’ve spoken about a lot as a firm and individually is responsible AI. It’s made you at times an outlier in Silicon Valley and I know there’s been fights with rivals including Andreessen Horowitz. Why is it such an important issue and what happens to you as investors, and also to these AI companies, if they don’t uphold that responsible approach? 

HT: If you think about the last technology cycle, with social media, we applied that technology to society in a pretty aggressive way. While it’s had a lot of positive impact, for the business models to work, social media also preyed on polarisation which had big consequences in society.

That is a global trend. You can go to every country and see polarisation and unrest, a lot of it caused by the fact that the technology cycle and the productivity gains that came from it were put in the hands of very few. It became a world of the haves and the have-nots.

If you think about what AI does, it takes that productivity and further concentrates it, into the hands of the very few. And this time it upends the labour productivity equation. If we’re not intentional about thinking about inclusive prosperity, then the people are going to be much unhappier than they were in this last cycle.

Capitalism is a privilege: it has to work for society or we won’t be allowed to practice it. Being careful about the role of applied AI in society is really important. 

GH: What could cost you that licence to operate? If these AI companies are in business at the privilege of society, what could cause that to be retracted? 

HT: If we eliminate jobs at scale, these service jobs, and we don’t reskill people to create a different opportunity for them, I don’t think that’s going to be a sustainable trend. I don’t see how that works. 

GH: You recently launched the General Catalyst Institute and this has been a conversation between [General Catalyst] and governments in various countries. What’s your sense of how the US administration is approaching the development of AI? Do you think Washington is taking it seriously and considering the ramifications? 

HT: I think administrations around the world are beginning to take this seriously and understand it’s an important issue. Businesses haven’t fully grasped it yet because we’re early in the applied AI cycle. We have to address this with urgency. 

GH: In the context of a race between the US and China for AI supremacy, is there a realistic shot at policymakers or companies limiting themselves in the race before something goes wrong? 

HT: The risk is that if we don’t help people understand their opportunities on the other side of adopting AI, we’re going to create friction, to create fear. 

It’s not just about “is our AI going to win versus the Chinese AI”, it’s also: will we create the conditions to accelerate the diffusion of this technology through our business that creates benefits across the entire economy. For that to happen, you need societal co-operation, which means people need to see the opportunity that they’re getting as a result of this as well. 

GH: How would you articulate that opportunity?

HT: The things that are important are how to make the work more interesting for people and, if AI is going to do functions x, y and z, what are we going to build on that to create further value? That mindset needs to be applied to every job function. 

GH: Are you beginning to see glimpses of human-AI collaboration where AI is augmenting human work rather than supplanting it?

HT: Absolutely. We’re seeing a lot of new innovations, a lot of founders working on this problem. One company where I’m very focused on this is Grammarly. Their purpose is to make humans more productive, to reimagine how AI can maximise human potential in the enterprise. 

GH: There’s a really interesting tension in this. On the one hand we’re talking about responsible AI and a rollout that respects human dignity in the workforce. Then, on the other hand, we have a hotter investment market almost than we’ve ever seen. OpenAI is leading the charge, but Anthropic, Mistral and other portfolio companies are a big part of that too.

How as an investor do you retain sobriety in that kind of environment? How do you stick to your values while also backing the fastest-growing companies and ensuring they can compete?

This is peak ambiguity and we’re all figuring it out in real time

HT: Let me just say this is peak ambiguity and we’re all figuring it out in real time. It comes down to backing founders with a true north. We have the responsible AI framework — the intention behind that has always been, if the founders can build with intentionality and give due consideration to the long-term impact on all the stakeholders, it buys us time to not be forced to regulate based on worries about what the technology can do.

It all goes back to: what is the responsible innovation playbook with which you’re building these companies? And if we as the innovation diaspora get that right, then it takes the pressure off of being forced to regulate fast, and creates market leadership in a way that’s sustainable and inclusive. 

GH: Is there a risk that a bubble is forming in the AI sector, particularly around the model companies? 

HT: Of course there’s a bubble. And by the way, bubbles are good. Bubbles align capital and talent in a new trend, and that creates some carnage, but it also creates enduring, new businesses that change the world. It happens in every bubble.

The thing that makes it peak ambiguity right now is, in the previous trends, you sort of knew what those technology trends were invoking and what the role of the companies driving those trends was. That’s what’s really unclear in this world of intelligence.

The trend is the models, then all this innovation gets built on top, both from an infrastructure perspective and also applications. But then the models get smarter and they gobble everything up, so which applications that get built on the models are going to be durable and what’s going to be subsumed by these models? That’s what makes it difficult to understand where to invest. 

If you think about some of the companies that got funded when [OpenAI’s earlier model] GPT-3 was launched, some of those don’t have a future any more because those functionalities got subsumed by GPT-4.

That trend has continued and you can see that with Anthropic and the great work it’s done in coding. So are we going to invest in companies that will fundamentally obsolesce because these models become more capable? That we just don’t know. We debate this every single day.

And that’s why we’re doing the roll-ups — we’re in effect acquiring customers and helping them be the beneficiaries of technology, no matter how it gets developed and transforming them — because that role is always going to be needed. 

And we’re also invested in companies like Anthropic and Mistral at the model layer where we think important companies need to be built.

GH: It’s an interesting question for start-ups in AI, whether their best outcome is getting acquired by an OpenAI, Anthropic, Cognition or Google, like Windsurf recently. There has been more consolidation for start-ups recently, does that change how you view valuations in these businesses, how you underwrite them, who you think can get off the ground?

You want to follow the lead of your founders who are close to the bleeding edge, and ultimately back their intuition on this

HT: It’s incredibly hard. These companies are growing very fast, so the valuations are extraordinary to reward them for the extraordinary growth, but what you don’t know is: is there durability or will they obsolesce? If you look at Windsurf, we were investors in that. I don’t know if the founders of that company are right or wrong, but their conclusion was that they will have greater impact in that industry by being part of Google and aligning the company with its own models.

Our belief always is, you want to follow the lead of your founders who are close to the bleeding edge, and ultimately back their intuition on this, so we were obviously supportive of that decision.

But it’s mind boggling. You would have never done that to a company like that in the previous cycles, you would be “wow, we’ve got a category winner”, with the way it’s growing and the way people use the product and how they love it.

That’s the kind of challenge we and some of the best investors and founders are feeling right now, in understanding what’s to come. 

GH: Well, it’s anathema to venture capital, isn’t it? If the best, fastest-growing companies decide that their best outcome is going to work at Google, that challenges the whole industry. 

HT: The other company that faces this is Cursor and they’re not doing that. They’re going at it. That’s why I said it’s more nuanced, one company has one theory of change, the other has a different theory of change. They are both excellent companies.

I don’t think the founders at these power law companies [the small number of investments that generate the majority of returns] are throwing in the towel. Look at Anthropic, OpenAI, some of these other businesses, that’s not happening. They’re all navigating ambiguity.

GH: One of the other things I wanted to touch on was sovereign AI [a nation’s own AI capabilities], particularly as you are a global investment firm. It’s increasingly a theme, given concerns around privacy and data usage and reliance on Chinese tech, but US tech increasingly for Europe as well. How much opportunity does that create for a company like Mistral in Europe or others elsewhere to go and build a sovereign model?

HT: Sovereign AI is not only a great opportunity for companies like Mistral, it’s also a great responsibility. You don’t want the service sector to melt into AI productivity for US companies, or Chinese companies. Figuring out a way to capture labour resources into AI productivity onshore is what will create the most resilient economies worldwide.

We need to help every region figure out how to do that or we’re going to hollow them out. Europe’s in a precarious position because their manufacturing got hollowed out to China, and if they also hollow out their productivity to the US, then it’s going to be very hard.

GH: You are investors in Mistral and Anthropic, two of the big foundation model companies. Do you think there is room for a whole panoply of foundation model companies or, as some other investors think, that this is going to be winner takes all in each region?

HT: In the end we’re not going to have 10 large language model companies with market-leading models and viable business models. You may have a few. A lot of people say there will be six, I think it’s going to be three or four. But then it comes down to where do they all specialise and focus? The sovereign AI opportunity will lead to local opportunities for AI models and infrastructure, given what’s at stake for those regions. 

GH: How much is that sovereign race affected by the combination of very mobile talent and high wages for the best researchers sparked by Meta buying [49 per cent of] Scale AI and throwing out nine-figure salaries for various people? If you can be anywhere and command that kind of salary as a top researcher, why would you not be in the US? 

HT: This comes down to AI researchers. Some can be mercenaries chasing the biggest dollars, some can want to work on the most interesting problems that might be afforded by some of these larger US companies and some might be chasing the patriotic local opportunities.

Some will not focus on making $100mn or $1bn or whatever the alleged rates are for the best researchers, but making a difference in their country or industry.

GH: Given everything we’ve spoken about — the money coming into the sector but also the ambiguity and the consolidation — where are you thinking about your potential exits in AI? 

HT: I really believe there’s going to be a whole wave of public companies that get created because of AI. A lot of the most exciting ideas in 10 years are going to be companies that are low-margin service businesses today that got transformed with AI. You’re going to see a lot of enduring businesses that get built there, you’ll see a lot in the model and the infrastructure layer as well. 

The M&A market is going to be vibrant as well not only to acquire talent, but also because, if the larger companies get some of those decisions wrong, they’re going to want to acquire the smaller ones to get back on track from a time-to-market perspective. It would be an interesting M&A market. 

This conversation has been edited for length and clarity.