Harvey: Product Launch

The leadership read

Harvey's 12x token growth in five months is less a product story than an infrastructure commitment story. At 1 trillion tokens a month, a legal AI platform can still be rationalized as a productivity tool with variable costs that track roughly with usage. At 12–13 trillion, the cost structure becomes a strategic liability unless the platform can route work intelligently across model tiers, matching inference cost to task complexity rather than defaulting to frontier compute for every request. Weinberg's public framing of this as the "billable hours problem" signals that Harvey has moved from adoption to justification mode: the question is no longer whether legal teams will use AI, but whether the economics of that use can survive scrutiny from GCs and CFOs. The related signals set for this period is thin on direct legal-AI comparables, but the cost-optimization pressure Weinberg describes is visible across adjacent verticals. Coinbase rerouting prompts to cheaper models and Uber questioning AI coding ROI are the cleaner analogies, both represent enterprises hitting the same inflection: volume growth outrunning unit-economics discipline. This is one of the sharpest articulations of that pattern from a vertical-AI vendor, rather than an enterprise buyer, in the last 90 days. Companies reaching this stage of inference-volume scaling face increasing demand for leadership at the intersection of AI infrastructure economics and commercial architecture, specifically, product and engineering leaders who can build model-routing and cost-attribution layers, and commercial operators who can translate token spend into client-legible ROI frameworks that hold up in procurement and renewal conversations.