I was a legal engineer before the Times named the job

The New York Times says the hottest job in law right now does not require a J.D. It is the legal engineer: someone who knows law, technology and business process well enough to rewrite how a firm works, then teach everyone else the new way.¹ I have been doing that job inside a personal injury firm for years. Nobody called it anything. Here is what it looked like on one workflow, and why a small firm should not hire for it or buy a tool that needs a translator.

TL;DR

• The Times reports that law firms, legal tech vendors and in-house legal teams are all hiring legal engineers, at posted salaries from about $124,000 to $320,000.¹
• Case launch at Conduit Law was 47 steps across six tools and about two and a half hours of staff time. It is now one click and about ten minutes, with a person approving what goes out. That rebuild is the job the Times just named.
• The vendors are hiring legal engineers by the hundred to help customers fit the product into their workflows. Read as an operator, that is a tell: the product needs a human translator per firm.
• A small plaintiff firm should skip both the hire and the translator. Stay in Claude or ChatGPT, put a firm brain on the systems you already run, and keep a person on every send.
• The first deliverable is a shorter software invoice. Fewer subscriptions, one consolidated stack, more cases without more headcount, stronger cash flow.
• That is the Managed Firm Brain: the legal engineer as a managed function inside our law firm operations work. The firm keeps its case system and its people.
• Request demo if you want the function without the headcount.

Case launch, before: 47 steps

I am not a lawyer. I founded Conduit Law, a personal injury firm in Denver that practices in four states, and I run its business side. When I first mapped how a signed client became an open matter, the map had 47 steps. Map the same job at most PI firms and you will land near that number, because the steps are not a Conduit quirk. They are what happens when a case has to be born in six tools that do not talk.

The intake form holds the date of loss, so someone retypes it into the case system, and the two copies drift the first time one gets corrected. Someone creates the matter, then builds the folder tree in the document server by hand, then fills the custom fields, then sets up the task list and the deadlines. The engagement letter goes out from an e-signature tool, the signed copy lands in an inbox, and someone saves it back to the folder. The retainer payment lives in another tab. The welcome email goes out twice, or not at all, because nobody is sure whether the matter is open yet. The team gets notified by whoever remembers, and the reporting row gets filled in at the end of the week if the week is quiet.

None of those steps is hard. Every one of them is a place to drop a client, and in PI a dropped client calls another firm. Two and a half hours of a good case manager's day went to clerical relay between tabs the firm already paid for.

Case launch, now: one click and a person who approves

The rebuild did not add a tool. It connected the ones that stay. The case system, the mail, the document server and the billing are wired into one layer, an MCP server we call the firm brain, that the firm's AI can read and write through. When a packet is signed, the brain opens the matter, builds the folder tree, fills the fields from the intake data instead of from a retyped copy, sets the tasks, files the signed packet where it belongs, and drafts the welcome email into the case manager's own drafts folder. The case manager reads it and presses send. Case launch takes about ten minutes, most of that the human reading.

That pattern has run across more than 1,000 cases. A person still releases every client-facing message, and live client calls still go to people. The read-first rule and the approval gates are not caution for its own sake. They are what let the brain take on the next workflow, because staff trust what it did with the last one. Records follow-up, provider letters and status updates came after case launch, one at a time. The parts we can share publicly are in the Conduit Law case study.

Look at what that rebuild required. Knowing the work well enough to see which of the 47 steps were the job and which were relay. Enough technical fluency to wire four systems into one layer. Enough process sense to decide what the AI may do alone, what it may draft, and what it may never touch. Then teaching a team to work the new way. The Times has a name for the person who does that now.

That was a legal engineer's job. The Times just named it

Sarah Kessler's DealBook piece ran October 31, 2026, and the reporting is specific.¹ Latham & Watkins is recruiting legal engineering attorneys in at least five cities at $220,000 to $250,000 and says more than 100 people now work on its AI initiatives. Kirkland & Ellis lists an AI innovation adviser at $153,000 to $220,000. Gibson Dunn lists a legal technology and AI attorney at $250,000 to $300,000. In-house, Nvidia posted a legal operations engineer at $124,000 to $195,500 and Microsoft a principal legal engineer at $147,000 to $278,900. Firm Prospects counted almost 50 lawyers who left the 200 largest U.S. firms for AI companies in the first half of the year, 15 of them into legal engineering titles. Stanford Law's two-week Legal Engineering Academy is not yet taking applications and already has about 1,400 senior leaders waiting to hear when it will.

Two quotes in the piece define the job better than the title does. Irene Liu, who runs Stanford Law's AI initiative, puts it in an equation: law plus technology plus business process. Mary O'Carroll, a legal operations executive who spent 13 years at Google, says the people who do this "don't have to have a J.D., and they don't have to be software engineers," and that the scarce skill is business process re-engineering.¹ That is the case launch rebuild in one sentence. The scarce part was never the code and never the bar card. It was knowing the work well enough to redesign it, and knowing the systems well enough to make the redesign real.

Why the vendors are hiring hundreds of them

The most interesting numbers in the piece are not the law firm salaries. They are the vendor headcounts. Legora has built a team of about 200 legal engineers since 2024 and hired about 70 in North America in the past year. Harvey grew its legal engineering team from 180 to 215 people in two weeks, all former practicing lawyers, at a posted range of $220,000 to $320,000. The Times says that at AI companies these people sometimes help build the product, but more typically help clients integrate it into their workflows, run trials and onboard employees.¹

Read that line as an operator. A product that needs one trained lawyer-engineer per customer to fit the customer's workflow is a product that does not fit the workflow on its own. For a thousand-lawyer firm with a hundred-person AI group, that may be a fair trade. The vendor supplies the translator, the firm supplies the budget, and the work gets done inside the vendor's portal. For a firm of eight to thirty people, it is a tell. You would be paying for the software and for a slice of someone's salary to make the software understand how your cases move, and at the end the know-how would live in a vendor's org chart, not yours.

The small-firm move is to skip the product and the translator. The frontier models the vendors are wrapping are already available to you directly. Claude and ChatGPT can both connect to a firm's own systems through MCP today, and both companies document how. What they cannot do on their own is know anything about your cases. That is the gap a firm brain fills: one layer on your case system, your mail, your files and your billing, so the AI you already pay for can read the matter, the folder and the thread at once. The home-surface argument is in live in Claude Cowork or ChatGPT, and the category problem with buying another legal AI tab is in legal AI software vs a firm brain.

Why a small plaintiff firm should not hire one either

If the work is real and the vendor route is a tell, the obvious answer is to hire your own legal engineer. Three reasons it is the wrong shape for a small firm.

The math. The posted bands are senior attorney money for a person who does not sign cases and does not settle them. Latham can spread a hundred-person AI group across thousands of timekeepers. A regional PI firm cannot spread one salary across anything.

The shape of the work. At Conduit the case launch build was intense for a few months. Since then the job has been maintenance, model changes, a new workflow each quarter and the occasional fire. That is a managed function, not a full-time desk. Hiring full-time for that pattern means paying for idle weeks or inventing work to fill them.

Key-person risk. The Times opens with Legora's Daniel Himmel remembering how nervous he was to leave practice in 2018 and how little he worries now, because vendors, firms and clients are all hiring. Good for him. Hard on the small firm that trains one. When the engineer takes the next offer, the prompts, the connectors and the reasons behind every design choice leave with them. Howard Glazer of Ropes & Gray asks the honest question in the piece: is this a transition position? For a firm with 100 people on AI, maybe. For a firm with one, a transition position is a single point of failure.

The Managed Firm Brain is the third option. FirmOps does the legal engineering as a managed service: the stack inventory, the consolidation, the build, the approval gates, the quarterly additions and the teaching. The firm keeps the brain. The firm also keeps everything else. FirmOps does not do case management, does not employ your staff and does not take over the back office, which is why it is not an MSO. That distinction is spelled out in law firm MSO vs a Managed Firm Brain. The brain is yours. The engineering is ours.

The first deliverable is a shorter software invoice

O'Carroll tells the Times that firms no longer need a giant pyramid of lawyers. At a plaintiff firm the pyramid that grew in the last five years is not lawyers. It is subscriptions. An intake tool, a texting tool, an e-signature tool, a document generator, a client portal, a CRM, each with its own tab, its own login and its own invoice, and each one bought to move data between the four systems that matter: the case system, the mail, the files and the billing. We mapped one such stack in cut a 10+ service stack down to a Managed Firm Brain.

So the first thing a legal engineer should hand a small-firm owner is not a roadmap. It is a shorter invoice. Keep the case system, whether that is Clio, Filevine or SmartAdvocate. Keep the mail, the files and the billing. Retire the tools whose only job was relay, because the brain now does the relay on top of the systems that stay. The outcomes that follow are plain ones: fewer subscriptions and a smaller software bill, one consolidated stack instead of tool sprawl, more cases handled without adding headcount, clients who hear back faster because fewer balls drop, and stronger cash flow because the work that moves a case toward settlement happens sooner. Your numbers depend on your stack, so we do not quote averages. We map them on the demo.

What to do this quarter

1. Map your own case launch. Count the steps from signed packet to open matter and circle every one that is relay between tools rather than judgment. The number will surprise you.
2. Write down every software subscription the firm pays for and the one job each one does. Mark the ones that exist only to move data between your case system, your inbox and your files.
3. Decide who approves what. Which drafts a person must release, which questions the AI can answer read-only, and which calls never leave a human.
4. Do not hire for it yet, and do not add another tab. Request demo. We show a live firm brain against that one job, confirm your systems have the APIs to build yours, and tell you whether it is a fit.

Questions owners actually ask

What is a legal engineer?

A person who combines legal knowledge, technical skill and business process design to rebuild how legal work gets done with AI, and to teach others to work the new way. The Times reports the title is used by law firms, legal tech vendors and in-house teams, sometimes as knowledge and innovation counsel, innovation attorney or legal applications engineer.

Do you need a J.D. to do legal engineering?

No. Mary O'Carroll says so in the Times piece, and the Times notes many legal engineers come from operations, paralegal work and consulting. I am a non-attorney running the business side of a PI firm, and the case launch rebuild described above is this work.

Why do legal AI vendors employ so many legal engineers?

The Times says they mostly help customers integrate the product into their workflows, run trials and onboard staff. Our read: a product that needs a trained translator per customer does not fit the customer's workflow on its own. A small firm is better off connecting the frontier model it already uses to its own systems.

Should a small law firm hire a legal engineer?

Usually not. The salary bands in the piece are senior attorney money, the work is not full-time in a steady way at a small firm, and the know-how leaves when the person does. Get the function as a managed service instead.

Is FirmOps a legal engineering staffing firm?

No. FirmOps does not place engineers at your firm. It builds your firm a brain on the systems you already run, then operates it and adds workflows over time. See what is a firm brain.

Is this an MSO?

No. FirmOps is not an MSO. It does not do case management, employ your staff or take over your operations. It makes the firm you already have more efficient: fewer subscriptions, one consolidated stack, more cases without more headcount.

Does the AI send things to clients on its own?

No. The brain reads first and drafts for review. A person releases every client-facing send, and live client calls stay with people.

How do we start?

Request demo. Bring your case launch step count and your list of subscriptions. We map whether a brain on your stack can carry it.

Sources

1. The New York Times, DealBook: The New Hot Job in Law Doesn't Require a J.D. (Sarah Kessler, October 31, 2026)