Here’s what’s on our play today:

• 🧪 Why law firms are borrowing engineers to build their own AI.

• 📰 Etched's $40B offers, Mistral beats China's models, Gemini's "Call for Me" expands.

• 🧠 Brain Snack: count how many people who can rebuild your AI actually work for you.

• 🗳️ Poll: does owning AI pay off for law firms?

Let’s dive in. No floaties needed…

The Laboratory

TL;DR

Big law firms are paying to own their AI, then borrowing the engineers who build it from AI companies.

• The launch: On September 17, 2026, Cooley released GO Public, a firm-owned filing tool built alongside engineers supplied by OpenAI.

• The floor: AI every firm can license gives clients no reason to choose one, so Kirkland & Ellis committed $500M to a platform shaped by 250 of its lawyers.

• The shortage: the lawyers who can build it command $200k to more than $400k, yet AI companies win them by paying in stock firms cannot offer nonlawyers.

• The workaround: about 35 of Kirkland’s roughly 50 dedicated AI engineers come from partner companies, so outsiders largely build what the firm owns.

• The stakes: no firm has published a return, so ownership stays a bet for a few rich firms until one employs more builders than it borrows.

Why law firms are borrowing engineers to build their own AI

In its early streaming years, Netflix filled its service with films licensed from Hollywood studios, which meant any rival paying the same fees could offer the same library. A library anyone could rent gave people no reason to choose Netflix, so the company began paying for shows nobody else could stream. In March 2011, it committed to two seasons of House of Cards, a political drama, before filming any episode. The show carried the Netflix name when it arrived in 2013, but an outside production company made it and kept ownership. Netflix had bought something exclusive, while the people who knew how to make television still worked for somebody else.

Big law firms are now in the same position with AI, because the legal AI on the market is licensed software that any rival can buy. A tool every competitor can rent gives clients no reason to choose one firm over another, so the largest firms have started paying to own AI of their own. Like Netflix, they are finding that the builders work elsewhere, so much of what they own is built by engineers AI companies lend them. The hiring race in recent headlines is therefore the visible part of a larger change, in which firms that always bought software are starting to own it.

The change became concrete on September 17, 2026, when Cooley, a firm of nearly 1.4k lawyers, launched a tool called GO Public. GO Public takes minutes rather than days to draft the filing a company needs before its IPO, and Cooley owns it even though its lawyers built it alongside engineers supplied by OpenAI. On September 24, the news agency Reuters (whose parent, Thomson Reuters, sells legal AI tools) reported that law firms were vying for technology talent, which is the hiring side of the same shift.

A licensed tool gives every firm the same floor

Firms pay for ownership because of how they make money, since a big law firm sells its lawyers’ time the way a consultancy does. A Georgetown Law and Thomson Reuters report found that 90% of legal fees are still billed by the hour, which sits awkwardly with a technology built to save time. Every hour AI saves is an hour a firm can no longer bill, so firms need their AI to offer clients something beyond the hours.

A bought tool struggles to provide that, because vendors such as Harvey build on models from OpenAI and Anthropic and sell the same product to any firm that pays. Jon Ballis, who chairs the law firm Kirkland & Ellis, told the Financial Times that such AI raises the floor for every firm, but “we don’t get hired for the floor.” A floor lifts every firm equally, so clients keep choosing on whatever sits above it, much as a rented film gave nobody a reason to pick Netflix.

Kirkland’s answer is to build above the floor, which is why it committed $500M over three to four years to a platform its outside builders may not sell to others. The platform is meant to work the way Kirkland’s lawyers work, because 250 of them, including 100 partners, are feeding into its design. The firm’s edge then comes from the partners’ knowledge the tool captures, which no vendor can sell to a rival.

Capturing that knowledge takes a ‘legal engineer’, usually a trained lawyer who can also build software. The job is turning how a partner structures a deal or drafts a filing into steps a model can follow. Because few people can do both halves of that job, every firm that wants to own its AI is hunting for the same scarce person, and so are the AI companies.

The richest firms borrow the builders they cannot hire

The competition shows up in job listings, where Bloomberg Law counted at least 16 top firms offering $200k to more than $400k a year for AI roles, most asking for a law degree. Even that pay loses to the AI companies, according to Andrew Detherage, managing partner of Barnes & Thornburg, who said anyone with those skills is being hired by Anthropic, Google, and OpenAI.

The AI companies win partly because they can pay in stock, which a law firm cannot offer anyone who isn’t a lawyer. The restriction comes from ‘Rule 5.4’, an ethics rule written by the American Bar Association and followed in some form by most states, which lets only lawyers own a firm.

The firms with the most money have responded by borrowing engineers from partner companies instead. At Kirkland, about 35 of its roughly 50 dedicated AI engineers are borrowed this way, much as OpenAI’s engineers helped Cooley build GO Public. As a result, the firms paying to own their AI rely on outside engineers to build it, which is the position Netflix was in with House of Cards. That arrangement assumes owning AI is worth what it costs, and not every firm accepts that.

Most firms license and train while a few build

Several firm leaders argue that licensing tools and training existing lawyers gets a firm most of the benefit, and Detherage’s own firm took that route. He says nearly 90% of its lawyers use AI, which suggests there is no industry-wide talent race, only a contest among a few very rich firms. That reading fits where the hiring happens but says less about direction, because firms that sit out still use the same labs’ models and stay on the floor.

A second objection is about money, since owning AI means paying for it whether or not it earns anything back. Daryl Lim, a law professor at Penn State Dickinson Law, made that case about Latham & Watkins, whose own NVIDIA servers pay off only if kept busy, because cloud providers spread costs across customers. The objection is fair but unproven, because no firm has yet published what its AI earns, which leaves ownership a bet placed by a handful of the largest firms.

Law firms own the platforms while AI companies employ the builders

Netflix eventually resolved its version of the bet by producing shows itself, which gave it ownership of what subscribers paid to watch. Law firms have not reached that point, since even Kirkland, the heaviest spender, still borrows most of its AI engineers. Borrowing also leaves open who owns what outside engineers help build, since the commercial terms of the law firm Freshfields' multi-year co-development agreement with Anthropic, announced in April 2026, are undisclosed. A firm that reports more AI engineers of its own than borrowed ones will have made Netflix’s move. Whether any firm can get there without being able to offer those engineers equity remains an open question.

Brain Snack (for Builders)

Quick Bits, No Fluff

• Etched fields $40B offers: AI chip startup Etched is reportedly fielding funding offers at a $40B valuation, a striking mark for a company betting on specialized transformer silicon.

• Mistral says new model beats China's: Mistral's CEO claims the lab's newest model outperforms leading Chinese systems in some areas, pushing Europe's case in the global model race.

• Gemini's "Call for Me" expands: Google is reportedly expanding Gemini's "Call for Me" feature, pushing its assistant further into making real phone calls on users' behalf.

Wednesday Poll

Meme Of The Day

The Toolkit

• Dust: No-code platform for building custom AI agents that connect to your company's tools and data, so teams can automate workflows without engineering.

• Krea: Real-time AI image and video generator with a creative-first interface, great for designers who want to actually steer the output instead of fighting prompts.

• Lavender: AI sales email coach that scores your drafts in real time and suggests rewrites to lift reply rates.

Rate This Edition