Key Takeaways

  • Harvey saw its gross margin collapse from positive 50% to negative 50% in June 2026 after agent token usage jumped 20-fold on frontier foundation models.
  • At an estimated $400 million ARR run rate ($33 million monthly revenue), a negative 50% gross margin created a burn of roughly $16 million per month purely on inference costs.
  • Founder Gabe revealed that Harvey returned to positive gross margins in a single quarter by introducing model routing and post-trained open-weight models (Harvey Tenant).
  • Flat-rate seat pricing collapses when customers switch from single-shot prompts to agentic reasoning loops that search whole firms and run recursive tasks.

The Agentic Token Explosion That Broke Fixed Pricing

For two years, software founders sold AI wrappers on standard SaaS subscription tiers. You charged a law firm $100 per attorney per month, passed their queries to OpenAI, and pocketed a 50% gross margin. Then came agentic workflows.

When lawyers stopped asking single questions and started letting agents run autonomous legal research across firm-wide document stores, token consumption went vertical. As John Coogan noted, “Harvey's gross margin fell from about 50% to negative 50% by June as agent token use spiked 20fold on rented OpenAI and anthropic models, Bloomberg reports.”

Selling fixed seats while paying per-token inference bills creates an uncapped liability. Coogan pointed out the sheer scale of the math: “So they had negative 50% gross margins in June. They're at a $400 million run rate for ARR I suppose and so that's $33 million a month and so negative 50% margins means they lost $16 million but they've raised like 500 million.” When your top customers extract the most value, they also consume the most compute, turning your best accounts into your biggest cash drains.

Fixing Unit Economics Without Forcing Premature Metering

When margins flip negative, the standard corporate reaction is to panic: hike prices overnight, restrict usage caps, or swap frontier models for cheap, low-grade alternatives. Harvey chose a different path.

Coogan highlighted the founder's strategy: “The easy thing would have been to force our customers into consumption pricing before they were ready and serve them worse models to protect our margins. We chose to help our customers transition on a timeline that works for them and give them the best models in the meantime.”

Instead of penalizing users, Harvey rebuilt its infrastructure layer. By routing queries dynamically and training specialized open-weight models (Harvey Tenant) for specific legal subtasks, they stripped out the bloated cost of querying external frontier APIs for routine lookups. “As a result, we improved our gross margins from negative 50% to positive in a single quarter despite usage doubling month over month and continuing to serve the best models.”

Comparing Harvey to legacy legal tech players like Thomson Reuters or CS Disco shows why this matters. Incumbents run on traditional software gross margins above 70%. If modern AI startups want to replace those giants, they cannot survive on raw model arbitrage. They have to control their inference stack.

What to Do With This

Audit your top 10% most active users this week. Calculate their exact monthly API cost against their subscription fee. If any customer is generating negative gross margins, implement an internal routing layer that diverts deterministic tasks to small open-source models before hitting proprietary frontier endpoints.