a16z Podcast

Jay Madheswaran on law's one real bottleneck: labor, not software

Jay Madheswaran· Co-founder & CEO at Eve
·~52 min·English·a16z
AgentsAI CompanyBusiness StrategyReasoning
TL;DR

Eve's CEO argues plaintiff law was never short on software — it was short on labor, so AI that removes that constraint doesn't cut jobs, it multiplies how many people a firm can represent.

01Core Mental Model

Labor Was the Real Bottleneck

<strong>Plaintiff law's one true bottleneck, Madheswaran argues, was never software — it was labor.</strong> Every problem the industry faced, it solved by hiring more people, so the tools that mattered elsewhere never touched its core.

No software really solved their core problem, which is labor constraint.

— Jay Madheswaran, a16z Podcast
Key Insight
Because labor was the only lever that ever worked, these firms learned to see every problem as something you fix by hiring — which is exactly why a tool that performs the labor itself, not just assists it, is a bigger break from the past than another productivity app.
02The Reframe

Revenue Expansion, Not Layoffs

<strong>For a plaintiff firm, AI is a revenue play, not a cost-cutting one.</strong> One Los Angeles firm went from about 100 clients per attorney to 500 — not by firing anyone, but by taking on cases it used to turn away.

And it's not because I'm cutting costs. It's like I can take on more cases. It's actually revenue expansion.

— Jay Madheswaran, a16z Podcast
Key Insight
The usual AI story — automate work, shed headcount — assumes demand is fixed. In plaintiff law demand is not the limit; the limit is how many of the waiting clients a firm can afford to serve. Lift that ceiling and the firm grows into demand that was always there.
03The Market Failure

David's Slingshot

<strong>The work a case needs, not the harm done, decides who gets a lawyer.</strong> If a case won't clear roughly $5,000 in fees, no one takes it — so most aggrieved people are simply priced out of representation.

If it's below the deductible, no lawyer in America will touch that case, right? It's just not economically worth it.

— Jay Madheswaran, a16z Podcast
Key Insight
This reframes access to justice as a unit-cost problem, not a moral one. The cases below the threshold are not frivolous — a $3,000 deductible can be a vanished paycheck — they are just too expensive to work by hand. Lower the cost per case and the justice system quietly widens.
04The Hidden Market

The Shadow Labor Market

<strong>Plaintiff law's tiny software spend hid an enormous one.</strong> Firms spend little on tools because they can't buy the thing they actually need — cheap labor — so the real market is the labor they'd hire if it existed.

So it turns out the software market is kind of like this shadow labor market and it's like massively bigger than anybody would have expected.

— Alex Rampell, a16z Podcast
Key Insight
Investors long judged plaintiff law by its software budget and found it too small to bother with. The budget was small for a reason — as Rampell puts it, the cheap labor these firms would gladly hire, a paralegal at 12 cents an hour, simply doesn't exist. Software that performs that labor collects against the outcome number, not the tool number.
05Why A Product, Not A Model

Einstein In A Law Firm

<strong>Raw intelligence alone would not be productive here on day one.</strong> Two things gate it: most of the relevant data never gets published, and every case turns on local context a general model has never seen.

If you drop Albert Einstein into a law firm today, he will not be productive for quite some time.

— Jay Madheswaran, a16z Podcast
Key Insight
This is the application layer's defense against being flattened by the labs. As the base models improve, Eve's value is not the intelligence — it's the private, outcome-labeled data and the encoded local playbooks that no amount of raw reasoning can reconstruct from the public record.
06The Adversarial Future

Agents Talking To Agents

<strong>When both sides get AI, the routine fight moves to machines and the human role narrows to judgment.</strong> Insurers will escalate too, so the edge becomes what the attorney uniquely knows about this client and this opponent.

I think a lot more cases are going to have teeth and all the routine back and forths I think is going to happen through agents talking to agents.

— Jay Madheswaran, a16z Podcast
Key Insight
A giveaway that this future is already here: one firm met a classic defense tactic — dumping discovery requests the day before a holiday — by building an Eve agent that answered the same day. The leverage isn't a smarter argument; it's never being the side that runs out of time or hands.
07The Business Model

Outcome Pricing Eats The Billable Hour

<strong>Software is moving from paying per seat to paying for outcomes — and contingency law has priced that way all along.</strong> Eve was outcome-based from day one because that is the only thing its customers actually buy.

What they care about is how do I serve more clients? So, that actually forced us to be outcome based from very early days.

— Jay Madheswaran, a16z Podcast
Key Insight
Outcome pricing only works when you can estimate, from a set of facts, how much compute a result will take — which is why it stayed rare. Eve's wager is that as long-horizon agents get reliable, that estimate becomes possible for more and more of law, and per-hour billing follows per-seat SaaS into decline.
08The Frontier

Ten Thousand Touches

<strong>A single case is roughly 10,000 back-and-forths over six months to a year — a long-horizon problem, not a single task.</strong> Eve isn't fully autonomous end to end yet, and the limit is engineers, not demand.

a single case has probably 10,000 back and forths.

— Jay Madheswaran, a16z Podcast
Key Insight
Framing a case as 10,000 steps explains why this is an agents problem, not a chatbot one, and why Eve's moat compounds: reliability has to be earned case-type by case-type, each one a small science of how to drive that specific matter to resolution. The bottleneck being engineers, not customers, is the tell of a market pulling harder than the technology can yet supply.