The economics, in law firm terms.
The price of this hire is public now. Bloomberg Law reported on 2 July 2026 that at least sixteen large firms had more than twenty-five AI strategy and capability roles open at once, advertised between $200,000 and $440,000. Pillsbury posted up to $440,000 for a director of data science and AI engineering; Latham & Watkins listed $295,000 to $400,000 for associate directors in AI governance. The reporting blames the trouble filling them on a candidate pool that has not kept pace with demand.
Those are large-firm numbers and a 45-attorney firm is not paying them. A 2026 recruiter salary guide compiling Built In, Glassdoor and Levels.fyi data puts senior AI engineers at $50,000 to $280,000 in base and tells hiring managers to add fifteen to twenty-five percent for bonus, equity and benefits. We budget the loaded figure at $185,000 to $280,000, the range we use on the general version of this comparison, plus a six-month minimum ramp before the first system lands. A firm bidding at the top of that recruiter guide's senior band should plan higher.
A commissioned build is a fixed from $10,000 for one named workflow, with a working prototype on the firm's real data in seven to ten days and production in four to seven weeks. One year of the loaded hire buys roughly one to three finished builds; two years buys three to six. The hire wins on arithmetic only if three to six builds are actually queued.
Firms have the money. The Thomson Reuters Institute and Georgetown Law 2026 Report on the State of the US Legal Market records 13 percent average profit growth in 2025 and talent costs up 8.2 percent over 2024. The constraint is not budget. It is whether the budget should become a permanent salary line.
What else that money buys.
Three other uses of the same money come up in the same partnership meeting.
A lateral hire. The strongest competing claim, and a revenue bet rather than a cost-structure bet. It carries a documented failure rate: research from Decipher Investigative Intelligence, reported by Above the Law in May 2026, puts lateral partner departures at 30 to 38 percent within five years. That does not argue against laterals. It argues the partnership already knows how to price a risky people bet.
A legal operations or practice technology director. For most firms in this band that is the better hire, and usually what the firm means when it says it wants an AI person. Salary.com put the US average for a director of legal operations at $244,992 as of 1 August 2026. That person runs vendor selection, adoption, policy and measurement. They do not build systems.
A commissioned build. A fixed scope against one named constraint, delivered and owned. A legal-ops director and a commission are complements rather than rivals: the director decides where to point the firm and holds adoption, the commission produces the thing being adopted.
The supervision problem no job posting solves.
Every other hire a law firm makes lands under somebody who has done the job. An associate reports to a partner who has run the same motion. A controller reports to an administrator who reads the same ledgers. An AI engineer reports to nobody who has managed engineering.
That is a governance issue before it is a culture one. The hire is a nonlawyer whose work product touches client matters. ABA Formal Opinion 512, issued 29 July 2024, builds on Model Rules 1.1, 1.6 and 5.3, the last of which covers a lawyer's responsibilities regarding nonlawyer assistance. The obligation sits with the lawyers holding managerial authority. It does not transfer to the technologist.
A firm that hires without an answer here inherits four unowned questions: who reviews the code, who decides the architecture, who signs off on the security posture, and who tells the partner group that the thing they asked for is a bad idea. A firm that cannot name a person for each is not ready to hire. It is ready to commission, where scope, acceptance criteria and review gates are written and signed before any code exists.
A commission does not remove the supervisory duty. It makes the duty reviewable, because the data boundary, the model selection and the human checkpoints are named in a document the firm keeps. That is the same paperwork a carrier or a client will ask to see, covered in our guide to bar rules and malpractice.
Where hiring is the right answer for a law firm.
Three to six builds genuinely queued. Not aspirations, a queue. At a firm that usually reads as new-matter intake and conflicts clearance, retrieval over the matter archive before a senior partner retires, billing narrative review, and client reporting. Four systems, four owners, four data boundaries. A firm holding that list in writing should hire.
AI as a position rather than a convenience. Some firms are pushed there by their own clients. Once corporate clients put AI sections in outside counsel RFPs and renewal packets, the answer becomes a capability the firm has to hold and keep current. Standing work wants a standing owner.
A partner who will actually sponsor it. The hire needs administrative rights on the document management system, access across practice groups, and enough standing to override a workflow three partners are fond of. Firms that hire the person and withhold the authority lose them inside eighteen months.
Where commissioning is the right answer.
One named workflow, not a roadmap. "Conflicts clearance takes us two days and we lost two matters this year waiting on it" is a build problem. Hiring a director to solve it is a $250,000 recurring answer to a $90,000 one-time question.
The firm would rather not compete for this candidate. The Bloomberg Law reporting is a warning as much as a benchmark. Sixteen of the largest firms are bidding, and technology companies are bidding against them for the same people. A 60-attorney firm in that auction should expect a long search and a hire who stays recruitable.
Ownership matters more than headcount. A commission ends with the firm holding the code, prompts, models, datasets, runbook (the written operating instructions) and integration documentation, inside its own cloud tenant (a private cloud account) under NDA (a signed non-disclosure agreement). No runtime to license, no per-seat fee. That is the same ownership outcome the hire produces, without the payroll line.
Key-person risk. When the internal engineer leaves, the knowledge leaves. When a commission closes, the documentation is the deliverable.
What we have built in law, and what we have not.
Systems built here have handled more than 6,000 live customer calls. The client we can name is Jim Glaser Law, a Massachusetts firm running five channel-specific voice agents across PPC, organic, TV, Meta and LSA, which have taken 3,787 calls across 5,514 minutes and give the firm per-channel attribution on every answered call. Jimmy takes reference calls for us. A prospect who would rather hear the work than read about it can call our own line at (617) 675-9067, which an AI receptionist answers.
A second law firm engagement we are not permitted to name is a platform build rather than a point solution: 13,296 matters, 4,396 clients and 5,684 invoices migrated off the firm's prior system, with the trust ledger reconciling byte-identical against the system it replaced at cutover.
What we do not have is a case study for every practice area, and we would rather say so here than in week three. If a firm's situation favors the hire, we say that on the call.