Key Takeaways

  • Software engineering represents a $500 billion labor market where AI can capture up to 50 cents of every labor dollar spent.
  • Legal AI platforms like Harvey and Legora target massive legal spend, but will likely capture only 10% to 15% of that budget.
  • Verifiability separates code from contracts: a compiler tells you instantly if code works, while legal text requires human judgment, context, and liability acceptance.
  • The radiologist effect applies directly to law: automating routine document review leaves the remaining high-touch, advisory work that justifies keeping human attorneys on retainer.

The Disagreement

When Jensen Huang declared that AGI had arrived with GPT Astra, Harry Stebbings saw an opening for enterprise software to swallow white-collar industries whole. Stebbings pointed to legal tech upstarts Harvey and Legora, arguing that if software development is a massive market, law must offer an equal prize.

“If coding is a half a trillion dollar market and you have two companies like Harvey and Lagora, I don't see why there's not a half trillion dollar market in law,” Stebbings argued.

Rory O'Driscoll pushed back immediately. The core difference between coding and law is not the total size of industry spend. It is the percentage of that spend software can replace.

“The only thing that mattered for the last two years is LLMs do code and code is a half a trillion dollar industry,” O'Driscoll noted. But that dynamic fails when applied to legal labor. “For every dollar you spend on labor you'll spend 50 cents on coding at least. In other words, that coding will do a lot of it. I think on legal it's about 10%.”

O'Driscoll pointed to medical imaging over the past decade to prove his point. Pundits predicted computer vision would eliminate radiologists by 2020. Instead, radiologist headcount grew. “The remaining quote 5% of the work turned out to be more than enough to justify 100% of the radiologists,” O'Driscoll said. “The grabag of tasks that make up law won't allow for the same percentage of total spend to move from human to AI.”

Who's Right (and When They're Wrong)

O'Driscoll is right about enterprise capture rates. Code is deterministic. You write a function, run unit tests, and the machine provides a binary answer: it compiles or it fails. Because the output is verifiable, engineering leaders feel comfortable shifting 50% of junior development spend to AI copilots and autonomous coding tools.

Law lacks that binary feedback loop. A contract draft is not simply right or wrong; it allocates business risk between two parties. The value of an elite partner at a law firm is not typing out standard indemnity clauses. Their value is signing their name to the document, carrying malpractice insurance, and advising a board during a crisis. An LLM can redline a non-disclosure agreement in three seconds, but that task accounts for a tiny slice of total legal billing.

Where O'Driscoll is wrong is underestimating the volume expansion. When the marginal cost of legal drafting drops toward zero, companies will not just draft the same number of contracts faster. They will run ten times as many contracts. Small businesses that currently skip formal terms or avoid trademarks will buy software to execute them. Legal AI will not replace 50% of existing law firm fees, but it will unlock a new category of autonomous legal ops for companies that currently cannot afford outside counsel.

What to Do With This

Audit your vertical AI roadmap this week. If you are building AI agents for white-collar workflows, calculate the percentage of determinative, verifiable tasks versus relationship-driven advisory work in your target job title. If deterministic output makes up less than 30% of the daily labor, pivot your pitch from labor replacement to workflow throughput, or price your product as an assistant rather than a full-time employee replacement.