Anthropic just dropped one of the most revealing studies on AI agents yet. Published February 18, 2026, the paper “Measuring AI Agent Autonomy in Practice” analyzed 998,481 real-world tool calls from their public API and Claude Code deployments.
The headline finding? Software engineering dominates nearly 50% of all agent activity today. Everything else — from business intelligence and customer service to high-stakes verticals — is still tiny.
Garry Tan, President of Y Combinator, saw the chart and immediately wrote one of the most important pieces for founders right now: “Half the AI Agent Market Is One Category. The Rest Is Wide Open.” (Published February 21, 2026 on garryslist.org).
His takeaway is blunt and actionable:
> “Software engineering owns 49.7% of agentic tool calls. Healthcare is at 1%. Legal is 0.9%. Education is 1.8%. A dozen other verticals are each under 5% — none above 9%. The red area on that chart is the biggest greenfield opportunity I’ve seen in years.”
Tan’s message to founders: Stop piling into yet another coding agent. The real unicorn factory is in the verticals that have barely been touched.
The Sectors Tan (and the Data) Say Are Wide Open
According to Anthropic’s domain breakdown and Tan’s analysis, these categories each represent less than 5% of current AI agent usage:
- Healthcare — 1% (patient records retrieval, risk scoring, clinical workflows);
- Legal — 0.9% (contract review, compliance, discovery);
- Finance (including crypto trading and regulatory reporting);
- Education — 1.8%;
- Customer service, sales, logistics, e-commerce, cybersecurity, and more than a dozen others — all in single digits.
Tan predicts this imbalance won’t last: “300 SaaS unicorns came before. 300 vertical AI unicorns are coming next.” And if he were starting a company today, he says he’d “stare at the red rectangular area of the bar chart until I saw my future.”
Why These Verticals Are About to Explode — With Hard Numbers
The broader market data backs Tan’s thesis hard:
- The global agentic AI market was ~$7.3–7.8 billion in 2025 and is projected to hit $52–140 billion by 2030–2034 (CAGR 40–49.6%), according to Fortune Business Insights, MarketsandMarkets, and Grand View Research.
- Healthcare AI agents alone are exploding: from $1.11 billion in 2025 to $6.92 billion by 2030 at a 44.1% CAGR (MarketsandMarkets). North America already holds 45% share, and multi-agent systems are growing at 45.3% CAGR. Adoption is already at 68% in some healthcare surveys.
- Financial services executives are even more bullish: 70% believe AI agents will directly drive revenue growth in the coming years (World Economic Forum + Accenture).
- Legal tech is seeing similar momentum — 87% of legal professionals predict AI will significantly transform the profession within five years (Thomson Reuters).
These sectors have three things software engineering doesn’t:
- Proprietary, high-stakes data (patient records, financial transactions, legal contracts).
- Heavy regulation that actually creates defensibility once you navigate it.
- Enormous economic pain points — a single AI agent that saves a hospital 10% on administrative costs or automates compliance for a bank can be worth tens of millions instantly.
Also read:
- Anthropic Just Shipped Its Own “OpenClaw” Faster Than OpenAI — Meet Dispatch for Claude Cowork
- Wasting Life: How Digital Existence Accelerates Time and Erodes Our Humanity
- 95% of Blockchains Aren't Blockchains: The Illusion of Decentralization in Crypto
Tan’s Practical Advice for Founders
Don’t just build another generic agent.
Tan says the winners will:
- Tap into proprietary vertical data;
- Master “context engineering” specific to that industry;
- Handle complex human–agent collaboration (most real sessions still need oversight — 73% of Anthropic’s tool calls have a human in the loop);
- Drive change management inside legacy organizations.
The autonomy numbers from Anthropic prove it’s ready: the 99.9th percentile Claude Code session length almost doubled from <25 minutes to >45 minutes in just four months (late 2025 to early 2026). Agents are getting dramatically better — but they’re still mostly fixing code.
The verticals are sitting there almost untouched.
Bottom line from the head of Y Combinator: The next wave of billion-dollar AI companies won’t be “AI for developers.” It will be AI agents that live inside hospitals, law firms, banks, schools, and logistics operations — places where today’s agent usage is still under 5%.
If you’re building right now, stop competing in the 50% red ocean of coding agents. Look at the tiny green bars on Anthropic’s chart. That’s where the next 300 unicorns are hiding.
