The most expensive lawyers in America are competing to build tools that will put themselves out of work. That, at least, is the faintly absurd premise of the legal AI boom. Harvey and Legora have raised more than a billion dollars between them selling software to the very firms whose hourly rates the software is meant to undermine. The pitch is efficiency at premium prices, a $1,000-a-seat tool for lawyers who already bill $1,350 an hour.
Talking Tree looked at that arrangement and saw something else entirely: an enormous population standing on the other side of the glass. The people who never make it into the lobby. The founder filing incorporation papers at midnight, the immigrant entrepreneur navigating a commercial lease, the rural shop owner who can’t find counsel within driving distance. For them, there was no tool at any price.
So Talking Tree built the tool, and priced it at what it costs to run. Founded in 2024, the organization is a 501(c)(3) nonprofit based in Sunnyvale, California, and its entire bet is that the most radical thing a legal tech company can do is sell its product without trying to profit from it. It offers AI drafting, contract review, and plain-English explanations at roughly 50 to 80 percent off what the market charges, and reinvests whatever comes in.
On paper it’s a mismatch. Harvey is worth an eye-watering $15.5 billion with a workforce of over 1,600 associates. Legora, at $5.6 billion, employs more than 400 people across six countries. Talking Tree has about seven, including a rotating cast of volunteers. But once you understand what’s actually being bought and sold in this market, the distance between what they can deliver turns out to be smaller than the headline numbers suggest.
The surprisingly level playing field
The first thing to understand is that all three companies are, in the most literal sense, running on the same engine. Talking Tree, Harvey, and Legora all build on the same underlying frontier models from OpenAI, Anthropic, and Google. None of them trains a fundamentally superior “legal brain” from scratch, and that single fact collapses much of the assumed distance between a scrappy nonprofit and a $15 billion juggernaut. When the models themselves are commoditized, raw intelligence stops being the point of differentiation.
*The one caveat is that Harvey is beginning to change this. It now trains its own post-trained model, Harvey Tenet, which it reports as a meaningful step forward, though the gains are self-reported on Harvey’s own benchmark and not yet validated by independent measurement.*
What separates the three companies, instead, is everything layered around the model. Harvey and Legora have poured their considerable funding into scaffolding that includes agents that run multi-step work across a matter, deep integrations with the systems lawyers already use, and heavy investment in the unglamorous work of making AI outputs more trustworthy. Harvey alone reports more than 25,000 custom workflow agents running on its platform. Talking Tree, by contrast, has built a leaner stack, an agentic legal companion named Redwood with voice support, contract drafting and review, an AI redaction tool, a template library, and a service that points users toward vetted lawyers. It’s a genuine product, but it doesn’t yet attempt the long-horizon, system-integrated work the incumbents specialize in.
None of this means Talking Tree’s reasoning is worse. It isn’t, and there’s no evidence to suggest it is. It means the company has consciously chosen to compete on a narrower band of work, the routine, high-volume drafting and review that dominates everyday legal life, rather than the complex, multi-step matters.
A founder’s odd advantage
Talking Tree Founding team (left to right): Michael Greenwald, Cecilia Wang, Alex Lu
Cecilia Wang is the rare founder who is also her own target customer. In private practice she did antitrust at Morgan Lewis, IP litigation at Quinn Emanuel, and before any of that, a federal clerkship on the U.S. District Court in Tennessee. Then she crossed into tech, first as corporate counsel in the semiconductor industry, then as Lead Counsel at Meta. As in-house counsel, she left Meta for a Senior VP and Deputy General Counsel role at a smaller public company and collided with the reality she’d later build a product around. Everything was limited: process, resources, time, headcount, plus what she calls “endless data chaos” that had a small legal team “working days, nights, and weekends just to keep up.” She built AI workflows to claw the time back and, by her account, cut her legal spend nearly in half. She still holds down an executive in-house role at Emerald Holding, Inc. (formerly NYSE:EEX), the events company recently acquired by Apollo (NYSE:APO).
Beside her are two co-founders who round out the picture. Michael Greenwald, the founding engineer, is responsible for building Talking Tree’s core: the AI pipeline, backend business logic, and the user interface. He’s a full-stack engineer who studied linguistics and computer science at UCLA, then spent a decade building across web, mobile, and the kind of emerging interfaces that don’t get much press. At the LA agency ISBX he built out back-ends and admin systems for startups and enterprises alike. He learned Unity in his spare time for the love of making games, and that hobby turned into a job at Stayhealthy, where he helped carry an augmented-reality health app called Color Quest past three million downloads, cutting its size by three-quarters along the way to keep it fast.
Alex Lu, who steers technology and infrastructure, has spent her career moving between art and technology, working in technical roles at Microsoft (NYSE:MSFT) and Intuit (NYSE:INTU) and doing visual design work at NASA’s Jet Propulsion Laboratory. Her past modeling stint, under the stage name “Alix Lu” (ig: @_alixlu), gave her chances to work with startups in a different capacity too. She produced investor-facing product videos and appeared as the public face of a makeup virtual-try-on startup, eventually being featured on the cover of Harper’s Bazaar Vietnam. All three graduated from UCLA. None of them draw a salary from Talking Tree.
That last detail is both the organization’s quiet strength and its most visible constraint. It’s a team that believes in the mission enough to work for nothing, which is exactly the kind of authenticity the nonprofit’s audience responds to. But it also means two of the three founders are holding down demanding full-time jobs elsewhere, and that the product’s day-to-day momentum rests heavily on Greenwald and a small supporting cast.
The economics of serving who no one else wants
Talking Tree’s LLM driven chat assistant tool
Where Talking Tree genuinely wins is in a place inaccessible to the giants. Harvey and Legora don’t just charge a lot, they also set minimums that exclude most of the market. A five-person in-house team signing with Harvey might pay for twenty seats, more than three-quarters of which sit empty. Talking Tree charges per actual user, from twenty dollars a month, a figure so small it looks like a rounding error next to Harvey’s nine-figure contracts. But that’s precisely the point. These are customers who were never going to buy Harvey or Legora, who were going to use nothing, or a generic chatbot, or a contract they found on Google.
The nonprofit structure is what makes that segment viable at all. Freed from the need to generate profit or maintain an expensive enterprise sales machine, Talking Tree can run at breakeven, and indeed it does. Roughly $427,000 in against $407,000 out in its latest public filing. Donations and grants help subsidize the mission, and revenue gets reinvested in accessibility rather than in a sales team. It’s an elegant model precisely because no profit-maximizing company would bother replicating it.
That same structural advantage, though, is also a ceiling. A small team can’t support the same hallucination-control engineering, security certifications, or agentic research as an organization with a $600 million war chest can. And there’s a subtler risk baked into its customer base. Every legal AI hallucinates, even the best. The big vendors hedge against this by selling to enterprises that employ lawyers precisely to catch those errors. Talking Tree sells to people making consequential decisions, forming a company, signing a lease, without a lawyer in the room. It’s a higher-liability position, and the organization is candid about it, routing users toward vetted counsel when the stakes rise. There’s also the squeeze from both sides of the market. Generic assistants keep getting cheaper and better at basic drafting from below, while the incumbents push down-market and a crowded middle fills in. The nonprofit pricing advantage gets harder to defend the moment a commercial tool can compete on both price and brand.
The temptation is to judge Talking Tree the way one judges Harvey, as a would-be competitor in a land-grab for legal AI. That misses what it is. Talking Tree lives in the corner of the market the giants walked right past. Its real invention was never technical. It was the decision to organize as a nonprofit and charge only what it costs to run, which let it serve customers whose combined spending wouldn’t register on either incumbent’s balance sheet. It also maintains a free newsletter with legal education, DIY guides, and plain-English explainers that anyone can sign up for at legalnonprofit.org.
The trade-off is plain enough. For the vast majority of buyers the giants ignore, Talking Tree offers the same model intelligence at a fraction of the price. What it gives up is the breadth of scaffolding and the depth of enterprise integration that all that venture money buys.
But Talking Tree doesn’t seem interested in out-building Harvey. It wants to make the legal system’s tools as accessible as Harvey’s prices are not. There’s something quietly defiant in that: a company whose entire competitive edge is deciding not to charge for one.