The largest companies of the coming decade will likely not be plain software companies, but AI-native services companies that use AI as their core engine to fundamentally redefine professional services such as tax, legal, and insurance. Making such a company succeed requires choosing a low-trust, outsourcing-based market, minimizing judgment at the task level, and securing rigorous quality consistency (variance control). Rather than acquiring an existing firm, the key strategy is to design from scratch for maximum AI operating leverage, capturing software-level margins and an enormous market size at the same time.
1. The rise of AI-native services companies and the new opportunity 🚀
The greatest companies to emerge from here are unlikely to be software companies that simply hand customers a tool. Instead, that position will be taken by companies that rebuild trillion-dollar professional services markets — insurers, law firms, accounting firms — from the ground up around AI, work that humans have primarily performed.
In the past, the dominant approach was selling copilot software used by the customer's own internal staff. Now, thanks to model progress, a new business model has opened up in which the company delivers the final outcome directly to the customer.
"Over the next decade, some of the biggest companies will not be software companies. They will be services companies — insurers or law firms — rebuilt from the ground up so that AI handles most of the work."
These companies operate in a completely different way from conventional tech startups, and they face explosive opportunity across vast professional domains: tax, audit, insurance, lending, healthcare, logistics, and more.
2. Four core conditions for picking the right market 🎯
When starting an AI services company, you should choose a field you can commit to for ten years or more. Markets where the AI services model can become explosively competitive share four distinctive characteristics.
- Low-trust markets: Domains where outsourcing is already the norm and the customer cares only about the final deliverable, not the process. You can absorb the existing outsourcing budget without needing to change customer behavior.
- Low judgment at the task level: When the whole job is broken into small pieces, most of the steps must be automatable. If every step requires subjective human judgment, scaling is impossible.
- A high intelligence threshold: Paradoxically, the overall difficulty of the work must be high. A barrier to entry only exists in domains where AI models combined with a small number of expert staff are required to produce output of a quality the customer will accept.
- The presence of regulation: Industries with strict regulation and legal liability make an excellent moat for a startup.
"You are not asking customers to behave fundamentally differently — you are replacing their existing outsourcing vendor. You step into a place where the customer's budget already exists and do that work for them, which is an enormous opportunity."
When evaluating a market, apply the so-called Sam Altman test: judge coldly whether your service becomes more powerful as foundation models improve, or whether the models themselves will replace (commoditize) your business. Also be careful with fields that require physical equipment or on-site labor, since it is hard to create software-style margin leverage there.
3. Three essential capabilities of a strong founding team 👥
The founding team of an AI-native services company needs a different combination of capabilities than a typical software startup.
- Domain fluency: Since you must deal with regulated industries and demanding enterprise customers, deep expertise and credibility are essential.
- Model fluency: You need technical leadership that accurately understands the limits and possibilities of frontier AI models and can evolve the product flexibly as models improve.
- Operational rigor: You must enjoy and be able to control operational concepts such as throughput, cycle time, and standard operating procedures (SOPs).
"The product is the operation. The vocabulary of operational metrics may not sound especially exciting to a founder, but you are fundamentally running an operations business."
As a real example, the team at General Legal, an AI law firm backed by YC, combined legal expertise from large law firms with legal-tech engineering leadership. They introduced a shift-based system for staffing attorneys, dramatically reducing turnaround time and maximizing operational efficiency.
4. The nature of product development and the existential threat of variance ⚙️
At an AI services company the interface facing the customer is not software but people (experts), and software plays the role of an internal tool that lets those experts scale non-linearly.
The most fatal problem here is precisely variance in output — inconsistency of quality.
"Customers will fire you far faster for inconsistent variance in the output than for being a bit slower or a bit more expensive than their existing vendor. Inconsistency destroys trust and drives churn."
- Identify bottlenecks and build internal automation tools to resolve them.
- Track throughput and cycle time as core product metrics instead of daily active users (DAU).
- Advance the system so that the human labor going into the service does not grow in direct (linear) proportion to revenue.
Early on you can run unscalable manual work alongside the system, but ultimately automating the process itself must become the core product.
5. Early sales pitfalls and the right pricing strategy 💡
The mistake early founders fall into most easily is the early demand trap. It is easy to gather pilot customers when a service launches, but taking on too many customers before you are ready means you lose the time to advance the product because you are patching everything with people.
So keep early pilot customers to a small number, observe their work deeply, and find the points where AI delivers differentiated leverage.
"Don't sell seats or tokens — sell outcomes. The pilot itself is the product."
Points to watch when setting prices:
- Recommended models: per-unit pricing (per tax filing, per loan review, and so on) or outcome-based pricing.
- Models to avoid: cost-plus pricing that adds a margin on top of cost, or straight-line undercutting on price.
Simply selling cheap can signal that the quality of the service is low, so price based on the value you deliver.
6. Proving AI operating leverage through the P&L 📊
The success or failure of an AI services company is ultimately decided in the structure of its P&L. Cost of goods sold (COGS) in particular must be managed rigorously.
COGS for an AI service consists mainly of three things: model costs, hosting costs, and the human labor deployed. The phenomenon in which human intervention falls as the product advances, lowering COGS and improving gross margin, is called AI operating leverage.
"Traditional services companies typically sit at around a 30% operating margin. But AI-native services companies aim, through AI operating leverage, to achieve software-like margins of 50% or more in a market two to three times larger."
Don't get addicted to pilot contracts with no margin or negative margin. Prove that operating leverage will work over the long run and that the business can evolve into a software-level margin structure.
7. Why acquiring an existing services firm is a trap ⚠️
Some founders with operating experience are tempted to acquire a legacy services vendor, layer AI on top, and generate revenue quickly. But apart from special regulatory purposes such as securing a license, this mostly ends in failure.
- Product-market fit cannot be bought with money.
- Legacy firms have completely hardened criteria for evaluating staff, organizational culture, and performance metrics, so simply introducing AI will not change their constitution.
- Designing the process around AI from the start and rebuilding from the ground up is far more effective.
8. Closing: preparing to seize an enormous opportunity 🏁
AI-native services companies represent a once-in-a-lifetime opportunity to transform professional services markets several times larger than the existing SaaS market.
Rather than getting lost in flashy features, establish an operational mindset that sees process as product and product as process, and control quality variance — and you can build one of the defining companies of the next generation.