“We built something incredible. And then we realized it was worth nothing to sell.” That’s how Ben Lasseri, founder of Plato, sums up the strangest side-project of his career.
In this episode of Build with AI, Victor Gross (Lead PM at Hexa) sits down with Ben Lasseri, founder of Plato, an AI copilot for litigation lawyers launched by Hexa just a couple of months ago. Plato’s first vertical is personal injury law (dommage corporel) in France. But the episode isn’t really about Plato, it’s about Plato JP, a jurisprudence search tool the team built on the side, in one week, without a single developer touching it.
The market nobody built for
Litigation lawyers who defend personal injury victims have one recurring problem: putting a number on compensation. French law breaks damages into roughly thirty separate categories (pain and suffering, loss of income, third-party assistance, and so on) and each one needs a defensible amount, usually anchored in precedent.
The existing legal-research tools, Doctrine, LexisNexis, Dalloz, are excellent at helping lawyers find a specific case. None of them are built to answer a statistical question: what have judges actually awarded, on average, for this specific type of injury? Nobody had built the numbers layer. The market was small too, around 2,000 personal-injury specialists in France, small enough that the big legal publishers had never bothered.
One week, Friday to Friday
Ben’s team decided to build that missing layer themselves, entirely with Claude Code, entirely outside their core product roadmap.
“We told ourselves we wouldn’t bring in a single developer. My CTO’s first reaction was: in two weeks they’re going to build me a rotten house of cards, and I’ll have to come clean up their code.”
The build ran in two phases. Week one: a working MVP, delivered Friday noon to Friday noon, with the team finishing at 1 or 2am most nights. Week two: polish, based on early user feedback, to make the product presentable to a notoriously demanding audience: lawyers.
The heaviest lift wasn’t the interface. It was the data. France’s court decisions are available as open data, but scattered across different formats and APIs for judicial versus administrative courts. Claude Code built a 12-step processing pipeline that scanned roughly 600,000 decisions and kept 80,000 clean enough to be useful.
“Two years ago, this would have taken a data scientist one or two months. It took us two days.”
The numbers that surprised even Ben
The tool launched quietly. It didn’t stay quiet for long.
Close to 1,000 lawyers requested access, roughly half of the entire French personal-injury bar. Around 500 signed up. Every week, about 150 log back in to search decisions. For a free side-project built in two weeks by a three-month-old startup nobody had heard of, that’s an unusually large chunk of a niche market showing up almost immediately.
The search and conversational layer runs on Algolia, still on the free tier, zero euros spent, plugged directly into an AI agent. It let a tiny team ship a search experience comparable to publishers who’ve worked the jurisprudence market for fifteen years.
“It’s worth nothing to sell”
Here’s the twist Ben is unusually candid about: Plato JP is a genuinely good tool, and it can’t be monetized.
“We understood we’d created something incredible. And that it was worth nothing. Any team could rebuild this in three weeks, a month. There’s no barrier to entry left.”
That’s not false modesty, it’s a structural read of what AI has done to software. The value used to live in the code: scraping government data, building search infrastructure, wiring up an agent. Today, a competitor with Claude Code and a weekend can replicate most of that. So Plato JP was never built to be sold. It was built to be given away, as a lead magnet and a product-discovery instrument, feeding Plato’s real, paid, RGPD-compliant product.
Turning credits into a sales pipeline
The business model reflects that logic directly. Users get 50 free credits, roughly 8 to 10 prompts a day. Once they run out, they’re prompted to book a call, where Ben pitches them the actual Plato product.
“Before, you’d write: ‘I’d love to show you our product.’ Now you give value first, for free, to the people who are your future customers.”
It’s a freemium play that would have been far more expensive to run before AI made shipping free software this cheap. And it doubles as market research: the team studied 200 to 300 users over three weeks, learning exactly how lawyers consume jurisprudence, insight that will shape Plato’s real jurisprudence feature down the line.
A media arm nobody had to hire for
An unplanned side effect: Plato JP produces weekly editorial content, also written with Claude. One piece analyzed gender bias in compensation awards and found that, across French jurisdictions, women are systematically awarded less than men for comparable injuries.
“This kind of editorial work used to take weeks. Today, it takes me two hours a week.”
Ben is quick to flag the catch: that barrier just dropped for everyone, competitors included. Proprietary-looking data and editorial output are no longer a moat by themselves, they’re a temporary edge that has to be reinvested constantly.
A house of cards, on purpose
Ben doesn’t oversell what he built. Plato JP will never become a Doctrine, he says flatly, it lacks the foundational data structuring, search architecture, and security work needed to serve hundreds of thousands of users.
“What you vibe-code lets you build a house of cards. Today’s house of cards is ten times bigger than what we could build six months ago. In two years, it’ll be ten times bigger again. But it’s still a house of cards.”
That’s a feature, not a bug, in his framing. The tool is a means, not an end, a “variable” in Plato’s go-to-market strategy. The day it costs more than it brings in, or the day the real product absorbs jurisprudence search natively, it gets deleted.
The bottom line
Plato JP is a small story with a large implication: when code stops being scarce, disposable products become a legitimate strategy, not a shortcut. Ben’s team didn’t try to build a moat with Plato JP. They built a two-week, zero-revenue tool, used it to learn their market and fill their pipeline, and stayed ready to throw it away the moment it stopped earning its place.
The lesson generalizes well beyond legaltech: as AI collapses the cost of shipping software, defensibility moves elsewhere, into proprietary data, network effects, or problems too specialized to be worth anyone else’s weekend. Everything else becomes a card in a stack you’re willing to knock down.
