The billable hour starts to give, and the seat-based model goes with it

At the Stanford CodeX Future of Law conference on 16 April, executives from Harvey and Legora, the two venture-funded legal AI vendors that now sit, between them, on about $16.5bn of paper valuation, told the room that their general counsel customers are, for the first time, writing fixed-fee alternative fee arrangements into their 2026 legal budgets. Harvey's CEO Winston Weinberg said client expectations have "already shifted". Legora's team said something similar.

This is a sentence that sounds like a conference platitude and is not. It is the thing the SaaS apocalypse frame predicts should happen next, stated by the vendors who are selling the tool that causes it, to the customers who are now buying on the basis that it does.

This is a sentence that sounds like a conference platitude and is not.

Let me walk through why I think this is a deal-mechanics story and not an adoption story.

The rounds

Harvey raised $200M in March at an $11bn post-money, per their own announcement and reporting in The Information.¹ Legora raised $550M in March at a $5.55bn post-money, up from roughly $1.8bn eight months earlier.² Two rounds, same month, both north of what the ARR disclosures, such as they are, would ordinarily support at conventional enterprise SaaS multiples. Harvey has talked about "over $100M ARR" in reporting through late 2025; Legora has been coyer. At $11bn on $100M-ish, Harvey is priced at roughly 100x. This is not a SaaS multiple. It is a priced-for-category-dominance multiple, which is a different thing.

The category-dominance thesis needs the category to exist as a large, durable revenue pool. And the pitch, made explicit at Stanford, is that legal AI's pool is not the software budget. It is the services budget.

What the AFA signal actually is

An alternative fee arrangement is, from the law firm's perspective, the thing the billable hour was invented to avoid. A firm quoting a fixed fee for a matter is taking on delivery risk, if the matter runs long, the firm eats the hours. Firms accept AFAs when they believe they can deliver the matter in materially fewer hours than the client would be billed for under hourly rates, and keep the difference.

AI productivity gains, if real, make AFAs rational from the firm's side: the firm produces the work in fewer hours and books the spread. They also make AFAs demanded from the client's side: once the GC knows the associates are drafting with Harvey, the GC does not want to pay for the associate hours that Harvey is doing. Both sides are pointed at the same structure. This is how pricing models flip, when both sides want the flip for different reasons.

The Weinberg quote, expectations have "already shifted", is the vendor telling the market that the flip is happening in contracted 2026 budgets, not in someday-pilots. That is, if true, the leading indicator that matters.

Where the frame fits

Run this through SaaS apocalypse. The frame says: per-seat pricing compresses as agents substitute for seat-holders. In legal, the "seat" is the associate hour, priced out through the billable-hour system. An AFA replaces a variable hour-count with a fixed matter-count. The associate hours don't disappear from the firm's cost base immediately, but they disappear from the pricing surface the client sees. Seat economics, as the client experiences them, compress to zero. The firm's internal seat economics compress later, as headcount plans adjust to the AFA margin.

This matters for Harvey and Legora's own pricing. Both vendors, as far as I can tell from their pricing pages and from reporting, are selling on a per-seat basis to law firms, seats for associates, seats for partners. If the firms they sell to are moving their own revenue off seats and onto matters, then Harvey and Legora are seat-priced vendors selling into customers whose revenue is de-seating. That is a slightly awkward place to be, and I would expect both vendors to quietly pilot matter-based or outcome-based pricing with their larger firm customers within twelve months. Watch for it.

Where the frame bends

The frame doesn't cleanly predict the valuations. If legal AI were simply a SaaS apocalypse story, seats compressing, you would price the vendors down, not at 100x. The 100x reflects a different bet: that Harvey and Legora aren't selling software into the legal software budget (which is small) but are taking a share of the legal services budget (which is enormous, US legal services alone is roughly $400bn). The AFA shift is the transmission mechanism for that value transfer. Services budget dollars, freed by AI, land partly with the firm (as AFA spread), partly with the client (as lower spend), and partly, this is the investor bet, with the vendor that supplied the productivity.

Whether the vendor actually captures that third share is the open question. Nothing in the Stanford remarks speaks to it. GCs signing AFAs into 2026 budgets is evidence that the value is being created; it is not evidence of who keeps it. Historically, enterprise software vendors selling into professional-services productivity have captured a thin slice. The 100x multiple assumes this time is different. It might be. The mechanism by which it would be different, weights-level lock-in, workflow entrenchment, data-network effects on matter outcomes, is not yet visible in either vendor's disclosures.

What this is a case of

This is the second professional-services vertical where AI-driven productivity is visibly pulling the pricing model off its historical base. The first was coding, Cursor, Cognition, and the seat-to-usage shifts at GitHub Copilot through 2025. Legal is following the same arc about twelve months behind, with the added wrinkle that the legal pricing model being disrupted is not a software pricing model but a services pricing model, which is ten to twenty times larger.

The pattern: AI productivity lands in a profession, the profession's billing model flips from input-metered (hours, seats) to output-metered (matters, outcomes), and the vendors that supplied the productivity either capture a share of the freed-up services spend or get commoditised by the model providers underneath them. The capture question is open in coding and it is open in legal. Consulting is next in line and will tell the same story with different nouns.

What to watch

• Whether Harvey or Legora shifts any published pricing away from per-seat in the next twelve months.
• Law firm disclosures, to the extent the AmLaw 100 discloses, of AFA mix as a percentage of 2026 revenue. The mix number is the scoreboard.
• Whether either vendor's next round is priced off a matter-volume metric rather than ARR. That would confirm the services-budget thesis is being underwritten by the investor base.
• Associate hiring at the AmLaw 50 for the 2027 class. Seat-count signal, one year out.

Footnotes

¶