Hello and welcome to Eye on AI. In this edition…Are services the new software?…Anthropic’s Mythos has financial regulators and bankers freaking out…more executive turnover at OpenAI…these measures may mean China may soon surpass the U.S. in developing the best AI models…are AI inference costs getting too steep?
Julien Bek never expected to go viral. Bek, who is an early stage investor in the London office of the venerable Silicon Valley venture capital firm Sequoia, says he merely wanted to highlight one of the firm’s recent investing theses and use the piece to highlight some of the startups Sequoia had recently backed. So he penned a blog with the title “Services: The New Software” and posted it to his social feeds. Within days, it had surpassed 1 million views on X. It is now closing in on 3 million. It has done more than 450k impressions on LinkedIn.
“I certainly didn’t expect to have this kind of reach,” Bek told me on a call earlier this week.
The provocative headline no doubt helped. But Bek’s thesis also struck a nerve. In short, he thinks that the world’s next $1 trillion company won’t sell hardware or software as a product. Instead, it will sell an outcome, and use AI-powered software to help deliver it, alongside human expertise. Instead of selling customer service software, for instance, it will simply deliver customer service for a client, the way business outsource processing companies do today. But these new entrants will be AI-native from the start. Instead of selling legal tech, these firms will sell legal services, etc.
Good examples of companies already pursuing this model that I’ve written about before include both Robin AI and Legora in the legal space and Dwelly in the real estate market. There’s also Dystyl AI in the consulting space, Rogo in financial services, and WithCoverage in the insurance brokerage market. Bek thinks there are many, many more to come. And he is sure the market potential is huge, noting that for every dollar enterprises spend on software, they spend six on services.
Intelligence vs. judgment
Bek has developed a taxonomy for thinking about these possibilities. First, he distinguishes between intelligence and judgment. Intelligence is basically anything with a pretty clear definition between the set of correct and incorrect answers—think tasks in coding, mathematics, physics, and even some tasks in accounting, law, or medicine. AI models are getting pretty good at delivering intelligence. Judgment on the other hand is more about taste, professional intuition, and subtle but often critical qualitative distinctions that often require both talent and experience. Lots of companies are trying to figure out how to imbue AI models with judgment, but for the most part, they aren’t there yet.
He then performs a matrix analysis that plots how a given service ranks on an intelligence-judgment scale on one axis, and whether companies already tend to outsource a particular service, or perform it in-house, on the other axis. (This is a complex decision governed by economic ideas that Ronald Coase developed in the mid-20th Century and that I recently wrote about in the context of the so-called SaaSpocalypse for Fortune here.)
First Bek looks at those tasks that companies already outsource to service providers, things like legal services, auditing, insurance brokerage, etc. Then he looks at the subset of those that are mostly about intelligence, with mainly just a dash of human professional judgment needed. This the sweet spot Bek thinks is ripe for AI-native service firms. “If [a customer] paid $100 for a service, but you offer them the same service for $80, but you can still do it at a high gross margin because you’re using a lot of AI to deliver that service, then we think that’s really interesting,” he says. Among the functions he sees being in this category are things like insurance brokerages, insurance claims adjustment, IT managed services, tax advisory services, accounting and audit services, simple legal services, payroll services, and certain compliance services.
Bek calls startups in this category—heavy on intelligence, with a dash of judgment, in categories that customers already outsource—”autopilots.” He says his use of that term in his viral essay has been the source of a lot of misunderstanding and misplaced criticism. He didn’t use the term to mean that services could be performed entirely by AI agents to the exact same standard as human experts. What he meant was that the processes that deliver these services could be largely automated in the same way that autopilots function in aviation—a human is still there monitoring the systems and handling the hardest tasks (like take off and landing) and ready to step in if something goes wrong, but a lot of the process is automated. He contrasts this to AI “copilots,” where he says there is a lot more back and forth between the human expert and the AI system.
I asked Bek about the theory that AI will enable some companies to in-source functions that they once outsourced. (That theory is part of what underlies the SaaSpocalypse—the idea that companies will choose to make their own software using AI coding tools.) He allows that this may be true for some functions, but insists that there are many things that will never be in-sourced either because of regulatory requirements—for example, financial auditing, in which companies must hire an independent firm—or for what he calls “softer” reasons. The latter category includes things like management consulting, which exists in part to provide external validation of decisions management already wanted to take—essentially helping to bolster their case to boards and investors, and, cynically, so that there is someone else to blame if it turns out to be a bad decision. The logic applies even in some IT functions. The old saying “no one ever got fired for hiring IBM” exists for a reason.
Not just a lower bill, a different bill
One of the biggest advantages the AI-native companies may have, Bek thinks, is around pricing. It’s not just that the AI-powered service firms can potentially charge less, they can charge differently. Many services firms in many sectors have billed by time. Billing by outcome changes the game completely. “When you’re a smaller company, the best thing you can do to compete with the larger ones is actually disrupt them on pricing,” he says. But he allows that it can take time to bring customers around to a different way of paying. For instance, people have been talking about getting rid of the billable hour in legal services for decades. The billable hour is, for most law firms that do corporate work, still here.
Bek insists that there are signs the billable hour really is going away, in large part thanks to AI. (There was movement in this direction before AI, but AI certainly seems to have accelerated it.) But, at the same time, he acknowledges that impediments remain. Some large companies use RFPs from services like consulting that ask for “an hourly rate”—if you price differently, you might not get past that screening because you can’t even complete the standardized form.
What about margins? One reason investors have loved software businesses is because they have often been extremely high-margin. Once you create the product, you can replicate it and distribute it at almost zero marginal cost. Anything based on human labor doesn’t scale the same way. Bek says the equation here is not as bad as some assume. In insurance broking, for instance, he says an AI-native startup like WithCover can sell 10x per human expert what a traditional insurance broker sells. “So I think the efficiency is proven, at least in some categories, not all,” he says. “But I think this is very encouraging.”
Two costs that are a potential issue: the cost of AI inference and the go-to-market costs of selling a service. Inference costs for running AI agents can, in some cases, eat up a substantial sum of money. (More on that in the Brain Food section below.) Bek cites figures from Bret Taylor, the CEO of Sierra, which sells AI-based customer service solutions, that gross profit margins probably look like 70% instead of 90% for some pure SaaS companies. But 70% is still a healthy margin. But the go-to-market costs remain an unsolved challenge, Bek says. You can’t scale enterprise service sales the same way you can software sales.
Sequoia isn’t the only investor with this idea. Private equity shops are betting that they can roll-up existing non-software businesses, infuse them with AI-driven efficiencies, and sell them off at much higher multiples. That’s why OpenAI and Anthropic both have major sales channels being built around private equity firms. But Bek thinks AI-native startups will be able to grab substantial marketshare faster than legacy firms can metamorphize into AI-first orgs.
He may be right. Change is hard. And having to reinvent both existing processes and existing business models is exponentially harder. The legacy companies have the relationships and the trust of existing customers. That’s often a trump card, especially for the highest-value work. But at some point, delivering an outcome at a lower price might tempt many to at least try the AI-natives.
With that, here’s more AI news.
Jeremy Kahn
jeremy.kahn@fortune.com
@jeremyakahn
But before we get to the news: Do you want to learn more about how AI is likely to reshape your industry? Do you hear insights from some of tech’s savviest executives and mingle with some of the best investors, thinkers, and builders in Silicon Valley and beyond? Do you like fly fishing or hiking? Well, then come join me and my fellow Fortune Tech co-chairs in Aspen, Colorado, for Fortune Brainstorm Tech, the year’s best technology conference. And this year will be even more special because we are celebrating the 30th anniversary of the conference’s founding. We will hear from CEOs such as Carol Tomé from UPS, Snowflake CEO Sridhar Ramaswamy, Anduril CEO Brian Schmipf, Yahoo! CEO Jim Lanzone, and many more. There are AI aces like Boris Cherny, who heads Claude Code at Anthropic, and Sara Hooker, who is cofounder and CEO of Adaption Labs. And there are tech luminaries such as Steve Case and Meg Whitman. And you, of course! Apply to attend here.
This story was originally featured on Fortune.com
This post was originally published here
