AI won't make white-collar expertise cheaper. It will make labor more accessible.
Technological Innovation
In the United States, white-collar services have always priced expertise at a premium.
We have seen technology break this pattern before. When information technology finally diffused through the American economy, the rate of productivity growth roughly doubled: total factor productivity growth accelerated from about 1% per year in the two decades before 1995 to more than 2% per year between 1995 and 2005. Notably, this payoff arrived long after the technology itself. Computers had been spreading through offices since the early 1980s. Economist Robert Solow famously quipped in 1987 that you could see the computer age everywhere but in the productivity statistics. Yet the gains only materialized once businesses reorganized around the technology rather than bolting it onto old workflows.
Then the surge ended around 2005 and productivity growth has been sluggish since. Robert Gordon and others use this to argue against extrapolating from the IT boom. But I'd argue the lesson cuts the other way: general-purpose technologies deliver their gains in waves, each one arriving only when the technology becomes capable enough, and cheap enough, to restructure an industry rather than merely assist it. Computing restructured the industries where the work was already digital. AI that can reason at scale is the first technology capable of restructuring the industries where the work is judgment: law, insurance, consulting, and above all, healthcare.
Historical Examples
PayPal is one of my favorite examples. Before PayPal, banks conventionally stayed away from small peer-to-peer transactions because every transfer required human review to guard against fraud: the unit economics simply didn't work. Millions of people paid a premium in fees and time just to move small amounts of money. Banks were not built to fight fraud at scale.
Around 2000, PayPal itself was losing millions of dollars per month to fraud, an existential threat at its margins. Max Levchin's team built detection software, internally nicknamed "Igor" after a Russian fraudster who had taunted the company, that scanned every transaction and routed only the ambiguous cases to human analysts (the episode is recounted in Eric Jackson's The PayPal Wars). Neither pure automation nor pure human review could have worked alone. The hybrid worked: fraud fell dramatically, and an entire segment of the population gained access to everyday digital money transfer for the first time.
Stripe followed the same pattern a decade later. Large banks had no incentive to make it easy for five-person startups to accept payments: serving that segment made no economic sense with the cost structure incumbents carried. Stripe saw a severely underserved segment and built the technology that changed the unit economics, turning an unprofitable market into a massive one.
The pattern in both cases: the incumbent wasn't stupid; the incumbent was rational. Serving the new segment was unprofitable given the incumbent's cost structure. The entrant won by building a fundamentally different cost structure, and in doing so created a market that hadn't existed.
Service Industries
As we enter a world where AI can reason at scale, companies built around deterministic software, and around expensive human judgment, face the same reckoning. BPOs, legal firms, healthcare systems, and insurance companies will be disrupted by AI-first models unless they reinvent themselves, which cash-cow businesses rarely do.
What's happening today fits a familiar historical pattern. Large enterprises (private equity firms, consultancies, law firms, hospital systems) are paying massive premiums to stay relevant. Incumbents acquire or adopt new technology and layer it over the broken systems they've already built, producing linear improvement. Meanwhile, entrants build AI-first from the ground up and improve exponentially.
Clayton Christensen, the Harvard Business School professor who coined the term "disruptive innovation," described exactly this dynamic, with a twist that matters here. In Christensen's framework, the disruptive product is usually worse at first. That's precisely why rational incumbents ignore it: it serves customers they don't want, at margins they can't stomach. The entrant gains a foothold in an underserved segment, improves relentlessly, and by the time it matches the incumbent on quality, it has already won on cost and reach. AI-first healthcare will follow this arc. It will start by serving the people the current system serves the worst, and it will not stay there.
Today
It's hard to categorize an entire market landscape in a few sentences without glossing over important details. But if I were to distill the models entrants are using to attack service-based industries, there are three: (1) a new era of consulting, (2) platform plus plug-and-play SaaS, and (3) full-stack platforms. All three are showing up in the market right now.
1) The new era of consulting. Mercor, one of the fastest-growing companies in Silicon Valley, recently launched an enterprise team that does full consulting work for service-based industries, charging a substantial premium for it. This is the highest-touch, fastest-revenue model: sell the transformation to the incumbents who are paying to stay relevant. Palantir coined this model: Forward Deployed Engineering. Another example is Ciridae. I got lunch with a Ciridae engineer about 6 months ago, and they are building an internal platform that allows them to horizontally integrate into service-based companies bought by private equity at a cheaper rate than what massive consulting firms would offer. And when I say cheaper, I don't mean $1,000 a month. I'm assuming they are still charging hundreds of thousands of dollars to customers for contracts, but conventionally speaking, a top consulting firm would charge a much higher premium and would most likely not have the technical capabilities to execute in this new environment unless they try to recruit AI engineers, who would probably have no idea where to start with the existing infrastructure of a large firm like McKinsey. So a true opportunity exists for new massive consulting firms to emerge.
2) Platform plus plug-and-play. Avoca AI is the sharpest example. Our team spent 5 months talking to hundreds of HVAC owners and going to conferences in the home service industry, and this company came up throughout (along with the hundreds of point solutions in the market). Avoca started with a niche problem (call handling and scheduling for high-volume HVAC, plumbing, and electrical companies) and plugged into incumbents like ServiceTitan, the most widely used software platform in the space. The wedge worked: Avoca has expanded its offering and in April 2026 announced more than $125 million raised across its Seed, Series A, and Series B at a $1 billion valuation.
Publicly, Avoca and ServiceTitan describe themselves as partners solving different problems: ServiceTitan as the system of record, Avoca as the conversation layer. My own prediction is that this partnership turns adversarial. If you talk to any home service business, they will most likely tell you the existing solutions are way too hard to use. The plug-and-play position caps Avoca's addressable market; owning the entire value chain is the only sensible way for them to win the market. But this is the danger of the model, too: attacking a software-first incumbent that owns the platform is challenging, because the platform can bundle, and it sits on the data moat. So it's not clear if companies like Avoca will win in the long term.
3) Full-stack platforms. A newer company in our own space, Lotus Health AI, has raised $41 million to date (a $35 million Series A co-led by Kleiner Perkins and CRV on top of its seed) to make primary care accessible to anyone. Lotus built its own electronic health record, recruited board-certified physicians onto an in-house clinical team, and acquired patients directly, currently offering care for free. Their bet resembles ours in many ways: asynchronous, AI-driven care is the future of accessibility. They are vertically focused on primary care, which some would argue is the right way to execute.
Healthcare
Today, every Shifu Health customer gets a licensed dietitian, a certified nutritionist, and a certified exercise coach. But the experience is roughly 90% AI: our AI health assistant handles intake, asks the right questions upfront to minimize back-and-forth, and manages the ongoing relationship. The human 10% engages when the AI flags a problem or the customer wants anything reviewed. We are actively recruiting across specialties (primary care, physical therapy, clinical psychology) toward a complete care team deployed on the platform, all in one app for the consumer.
The question driving us: how do we make preventive healthcare more accessible? A large number of Americans are on mediocre health insurance that prevents them from getting holistic preventive care today. I plan to dive deeper into this specific topic in another essay.
You probably know someone in your life who would benefit from an on-demand health team that could be there to help with anything health-related at whatever time they needed. The reality is that to get this in America, you have to pay a premium most people cannot afford. What if you eliminated this barrier to access? That's why we started Shifu Health.
If you'd like to talk more about the way we are approaching the problem of unaffordable healthcare in the US, you can email me at spencer@shifu.health.