Can AI replicate an army of collaborators? These lawyers bet their new firm on it

Sam Shaddox and Matt Souza spent years in the long-term legal profession, as attorneys at a large Seattle firm and later as general counsel at technology companies. They watched as law firms charged start-up clients exorbitant fees for work they believed AI could do quickly and cheaply.

Talairis Law Group is their answer. The Seattle-based company, which launched this week, is built on the idea that AI can handle much of the work done by big law firms — and that startups shouldn’t have to pay huge legal fees for it.

“It’s the start of this AI moment,” Shaddox said, “and it’s the start we all need and the Seattle startup space needs. We’ve been on the other side of the aisle, and now it’s time to make our mark.”

The idea is no different. Spending has poured into AI law firms over the past few years, with players like Crosby, Manifest OS, Eudia and Lawhive raising hundreds of millions of dollars combined. But Shaddox and Souza say those firms have chosen one practice area — contract review, immigration, M&A due diligence — that leaves a gap that no one has filled.

“They’re all choosing the same path,” Souza said, “and there’s no AI-powered law firm that you can rely on to help you with your day-to-day stuff as things come up, helping to steer your ship.”

Founders: Shaddox and Souza were both associate attorneys at Perkins Coie, a prominent Seattle law firm, before moving in-house to Seattle technology companies.

Shaddox went on to legal work at Big Fish Games and OfferUp before serving as general counsel at SeekOut, an AI-enabled intelligence firm. Souza was a senior advisor at Zillow before becoming general counsel at Wrapbook, an entertainment income and financial platform.

They say it was that experience at home that made this problem impossible to ignore.

“We were being bombarded with work that — as we were adopting AI in-house — we saw law firms that weren’t doing well, or weren’t doing well,” Souza said. “The whole economic model for law firms is broken. So that’s where we started.”

How does this work: Talairis is built around what the founders call a four-layer architecture:

• Underneath is the main language model — the AI ​​engine.
• On top of that sits what they call the agent layer, with more than 100 purpose-built AI agents that cover the range of legal tasks a startup might need.
• What’s more is what they call the “client genome” — a stored profile of each client’s business, risk tolerance, contracts and performance history, so advice is never ineffective.
• And at the top are Shaddox and Souza themselves, who review and sign off on all deliverables.

“You don’t get just one piece of advice that doesn’t know what your company is or does or how it thinks and works,” Shaddox said. “You get unexpected results.”

By working: As an example, Shaddox and Souza point to SAFEs: simple future equity agreements, a common instrument for financing the first bridge. First-time founders often try to handle it themselves, or bring in outside counsel for $1,500 an hour. Either way, manually working with notes, side characters and cap table effects is painful and error-prone.

Talairis is designed specifically for the agent. Send them SAFE, they say, and you get back more than a legitimate idea.

“They don’t just come back, ‘Hey, here are our thoughts on this convertible note’ — anyone can do that,” Shaddox said. “Instead, they get a completely structured cap table that includes the latest note, includes the side letters, and shows how that will go in their next earnings.”

The first step: The company is being launched by paying customers, although Shaddox and Souza are not naming names yet. Talairis has been arrested and there are two lawyers at the moment.

• Pricing: Shaddox says their hourly rate works out to about half that of a big legal attorney, and that the AI ​​multiplies the output enough that the effective cost to clients is a fraction of what they’d pay elsewhere.
• Privacy: To the question of whether client data is used to train AI models — a real concern for startups sharing sensitive legal documents — Shaddox is specific: “The answer is no. Your data is never used to train a model.” Talairis has built privacy and attorney-client privilege protection into its design from the ground up.

The launch comes the same week Anthropic released Claude for Legal, a collection of more than 20 new connectors and plugins for 12 practice areas aimed at bringing AI tools to law firms and in-house legal teams. Shaddox sees time as validation.

“Claude for Legal and any other LLM is fundamental,” he said. “Our unique approach is the one that sits at the top: a law firm with top lawyers, key proprietary enhancements, scope for each client, rights protection, and an agent structure that lacks a standard plugin. That’s what makes LLM out of the box the best legal advice to start.”