Teamday
← Back

Eve's Jay Madheswaran: AI Lets Lawyers Take 500 Clients

a16z
agentsknowledge-workbusinessstartupinterviewfuture-of-work

How Eve Turns Plaintiff Law Firms Into AI-Powered Businesses

Jay Madheswaran is the co-founder and CEO of Eve, a legal AI platform built for plaintiff law firms: the contingency-fee lawyers behind personal injury, employment and consumer cases, rather than the corporate firms that most legal software targets. Eve raised a $103 million Series B at a $1 billion valuation in September 2025. In this a16z conversation with general partner Alex Rampell, Madheswaran explains why plaintiff firms are labor-constrained rather than demand-constrained, what an AI agent does across a case that takes months, and why Eve charges for outcomes.

Labor, not demand, is the bottleneck: Plaintiff lawyers get paid only when they win, so they turn away cases that can't cover the work. "Today, if a lawyer doesn't make at least 5,000 fees off a case, they do not take it on." Madheswaran's example is employment law: screening one caller takes 30 to 45 minutes, followed by around 10 hours of legal research to tie the facts to a cause of action, and 96% of labor and employment cases end up rejected. Eve answers the intake call, qualifies the case against the firm's own criteria and starts the preliminary work, which changes which cases are worth taking. Rampell's framing: the plaintiff software market looked tiny because firms barely bought more than Microsoft Office, but the real market is the labor they couldn't hire. Software is a "shadow labor market."

100 to 500 clients per attorney: The headline case study is a Los Angeles firm. "One firm in LA went from managing about 100 clients per attorney to 500." The firm used Eve from intake and qualification through the handoffs between attorneys to settlement, where Eve prepares mediation briefs that suggest a target outcome and the tactics that worked on similar cases. Madheswaran says this is revenue expansion, not cost cutting: firms take more cases. An Eve product called Auditor reviews every page of every case overnight for issues a perfect attorney would catch, and he says almost every case has missed information.

Why raw model intelligence isn't enough: "If you drop Albert Einstein into a law firm today, he will not be productive for quite some time." Fewer than 1% of plaintiff cases become public opinions. Most settle before litigation, so knowledge such as how to write a demand letter to a given insurer for a soft tissue injury exists nowhere online. Eve never trains on privileged client data. It mirrors how attorneys already share tactics at CLE events, employs its own attorneys and tracks case success rates across the platform.

10,000 steps per case, not yet fully autonomous: Madheswaran estimates a single case involves around 10,000 back-and-forths, from requesting medical records (which takes two weeks today) to negotiating with the other side. Eve's outbound voice agents now make those record requests, and he says agents are already talking to other agents and learning how to get past each other. Asked whether Eve has run a case fully end to end, he answered: "Not yet, but we're close." The constraint is engineering capacity, not demand: "we're largely limited by engineers at this point."

Outcome-based pricing and firm economics: On a $25,000 settlement, the attorney takes about a third, roughly $8,000; about half of that goes to labor, 30% to marketing, and the rest is profit. Eve started with tiered seat-style pricing, then moved to outcome-based pricing about a year and a half ago because firms care about serving more clients. Madheswaran expects law firms to "look more like software companies": specializing in narrow case types, building their own data sets, and in some cases rolling up other firms because Eve shortens post-acquisition integration.

The defense side is still billing by the hour: One customer kept receiving discovery requests the day before Thanksgiving and other holidays, a tactic meant to ruin the plaintiff team's vacation or rush the response. The firm built an agent on Eve to respond the same day. Madheswaran expects routine back-and-forth to move to agents talking to agents, with attorneys spending more time on trial work, and some disputes moving to AI-assisted mediation outside the courts.

  • AI as revenue expansion - Plaintiff firms are supply-constrained, so cheaper labor means more cases taken, not fewer people
  • 100 to 500 clients per attorney - One LA firm used Eve end to end, from intake to settlement briefs
  • Proprietary data beats raw intelligence - Settlement knowledge isn't public, so Eve's value is context models can't scrape
  • Outcome-based pricing fits outcome-based customers - Eve dropped seat-style tiers once it saw firms only care about serving more clients
  • Long-horizon agents are the frontier - A case has around 10,000 steps; Eve says full autonomy is close but not yet live

What Eve Means for Vertical AI Agents

Eve's case shows where vertical AI agents earn the most: industries where demand already exists and labor is the limit, so every task an agent absorbs turns into new revenue. Plaintiff law was always paid on outcomes, which made it a natural fit for outcome-based pricing, and Madheswaran's bet is that other professional services will follow. For organizations deploying AI agents, the lesson is to target the bottleneck that caps revenue, not the line item that looks expensive, and to accumulate the private, domain-specific data that general models don't have.