Legal departments adopted artificial intelligence ahead of most of the firms that serve them, on flat budgets with small teams. What follows is how to turn that head start into method, and method into leverage.
Enclosed is your invitation to a private forum: a small circle of in-house leaders, lawyers and business people alike, who accelerate the learning, the connections, and the capability that no department builds alone.
Your lawyers using artificial intelligence themselves. Drafting, summarizing, preparing. Each person faster on their own work.
Intelligence inside the vertical tools you already buy: billing review, contract analysis, research. Both are real progress. Both stop at the same line.
Agentic workflows that run end to end, with your lawyers supervising by exception rather than touching every item.
Real progress. It speeds up a person.
Real progress. It speeds up a product.
The third form has to be engineered around how your department actually works.
And you have no billable hour to protect. When artificial intelligence takes an hour out of a workflow in a law firm, that hour was revenue, and the firm has to decide who absorbs the loss. When it takes an hour out of your department, the hour goes straight back into work you could not get to. You carry the professional risk of this technology. You do not carry the economic threat. That asymmetry is why in-house teams are moving faster than the firms that serve them.
Brainyacts Research · First Findings · July 2026 · in-house respondents n=29
One. The work you already rely on, the summaries, the first drafts, the triage calls: does it hold up, edge cases included? Not in the demo. In production, on the matters where being wrong costs something. The moment a workflow runs end to end, that question stops being optional.
Two. Your firms are adopting the same technology. Is their efficiency reaching your fees? In our First Findings survey, 55 percent of legal departments said they have no way to estimate how often a firm used AI to do the work faster without a matching reduction in the bill.
Each card opens. Click any one to read the section behind it.
Someone has to decide where AI belongs inside legal work, and prove it holds up.
How the work actually happens today: people, systems, exceptions, and the routing logic written down nowhere.
Where AI belongs, and where it does not. What stays deterministic, what gets a model, what stays human.
Working software carrying real responsibility: an audit trail, evals, and escalation to a lawyer built in.
This is the person who takes a department from the first two forms of adoption to the third.
The copy that lands on legal's desk every week, product claims, sales decks, pitch materials, campaign copy, each piece checked for regulatory exposure, claims substantiation, competitive statements, and IP flags. High volume, real judgment on every page. The stated workflow is five clean steps. The real one lives in the exceptions.
Click any step to open it. The clean version hides the real work.
The instinct is to look for one person who holds all of it: a lawyer who has worked in the trenches and above them, who reads process and technology and the whole enterprise, with the vision to see the workflow and the legal need at once. That person is real, and almost no department has them on staff. Here is the reframe that matters. You do not need that person. The hiring market for this profile is thin, expensive, and aimed at firms and vendors with money to burn. A legal department runs lean and does not have the headcount for a new position, and it does not need one. The scarce half of the role is judgment, which your people already have. The engineering half is teachable, to lawyers, to legal ops professionals, to the other professionals already on your team, and it gets easier to learn every year.
Brainyacts Research · First Findings · July 2026 · in-house respondents n=29
And it is not a step down. The lawyer who learns this stops being one more pair of hands and becomes the person with the clearest view of where the work is going. It is fast becoming a path to the general counsel's chair, because the next GCs will be the ones who understand, first hand, what artificial intelligence can and cannot do in legal work.
This loop is forward deployed legal engineering, and it is the only way the third form gets built. Agentic workflows are not bought finished. They are learned into place: you embed to see how the work really happens, you run evals to earn reliance, you deploy under supervision, and then you go around again, because the work drifts and the models move. Learning in loops is what separates meaningful adoption from the performative kind. It is what makes the work defensible, and what lets you rely on it responsibly. And one phase, evals, is about to feel very familiar.
learns how the work really happens and finds the workflow worth rebuilding.
prove the system behaves, with evidence you could put in front of a board or a regulator.
makes it work inside the department, on your own systems.
Sitting with the work determines what should be built before anyone builds anything. It ends in one artifact the whole department can argue with.
swipe the diagram sideways if it is cut off
Prioritize high volume workflows where the improvement is large enough to matter. Then design the split.
Most of the workflow should not be AI at all. Deterministic software does what it is told. The model gets exactly the steps that need judgment. A lawyer holds the decision. And every step, human or machine, lands in the audit trail.
The forward deployed legal engineer understands the current workflow, designs the intelligent one, and builds the system that connects the two.
You already run diligence on everything you rely on. Evals are diligence on the machine's work: a golden dataset of real cases, labeled by your lawyers, run against the system, with the evidence kept. Reliance on AI in legal work is only as defensible as the record behind it. This is that record.
| Right context | Rule applied | Matches the lawyer | Safe to clear | |
|---|---|---|---|---|
| Standard product claim | ✓ | ✓ | ✓ | ✓ |
| Comparative naming a competitor | ✓ | ✓ | ✓ | → lawyer |
| Nested disclaimer in a footnote | ✗ | · | · | → lawyer |
| Regulated efficacy language | ✓ | ✓ | ✓ | → lawyer |
| A claim type not seen before | ✓ | ✗ | · | → lawyer |
swipe the table sideways to see all columns
Each failure names a category the agent cannot yet handle, which becomes an escalation rule and the next round of the loop. The same loop does something quieter and more valuable: it moves your team's past decisions, inconsistent and human as they are, toward one standard the general counsel approves. The prize is not only an agent that works. It is more consistent legal judgment across the department. Numbers illustrative.
swipe the diagram sideways if it is cut off
Build on the systems you already run: the contract tool, the matter system, the DMS. Never force a migration.
A sandbox on your own licenses, to run, inspect, and debug safely. No privileged data leaves.
Shadow mode first. The smallest useful action. More authority only as the evals prove it.
The eval results and the running audit trail are the file you could put in front of your board, or a regulator, and stand behind.
You hold the most powerful position in this entire picture, and the data says almost nobody has pressed it yet.
Brainyacts Research · First Findings · July 2026 · in-house respondents n=29
The firms feel it coming. In the same research, 69 percent already field client questions about AI and the bill, yet 41 percent have made no change to their rates, structures, or guidelines, and only 10 percent have a formal policy. The conversation has started on both sides. It has not reached a single invoice.
That is not a grievance. It is an opening. You are the buyer, and the standard is yours to set. It is the standard you just ran on your own work, pointed outward. Two demands, to every firm on the panel.
The instrument already exists in your world: the outside counsel guidelines. Add one page. An evidence addendum, framed as questions any firm using AI on your matters should be able to answer.
No firm can call those questions unreasonable, because they are the questions you answer internally every quarter. One rule makes the whole section work: you cannot demand a standard you do not practice. That is why the loop comes first, and why the department that runs it earns the right to ask.
Between us, more than twenty years across every era that led here. Each one taught the same lesson from a different angle: the technology is never the hard part. The work, the incentives, and the people are.
Strategy counsel to global law firm and in-house leadership on pricing, growth, and business model design. He has built books of business past $20 million by redesigning how firms price, sell, and deliver, and has stood up captive delivery arms and new practice models inside them. Professor of legal innovation who founded the University of Richmond's Legal Business Design Hub, a Fast Company 2022 Innovation by Design winner. Publisher of The Brainyacts, read by thousands of senior legal professionals.
Technology founder who built a legal software company and sold it to Thomson Reuters. Two decades of systems inside and alongside the Am Law 100: portals, platforms, knowledge systems, and now agentic AI. Six Sigma Black Belt. Fluent in both firm economics and software architecture, which is rarer than it should be.
Every era taught us what survives contact with a legal department. This is the first one where the tools can keep up.
Some general counsel will grow a forward deployed legal engineer on their team this year. All of them will need to think like one: to ask for the map before the tool, the eval before the rollout, the record before the reliance. The fastest way to make it yours is a working room: general counsel, in person, learning the method and doing the work on a live workflow.
Four hours, one room, in person. A working session, not a webinar: half of it is learning the method, half is doing the work, mapping a real workflow live and drafting the evidence addendum together. You leave with sharper language, working tools, and a concrete way to take your department's AI adoption deeper.
There is no ticket. Admission is by contribution to the research behind The Legal Economics Study 2026, our independent benchmark of what AI is doing to the cost, price, and value of legal work. Take the deeper survey before the room, sit for a confidential interview after it, or host a future session. No vendor sponsors it, and no finding is for sale. Contributors see the findings first.
The vocabulary. The reliance frame. Your first workflow map. A draft evidence addendum to your outside counsel guidelines. And a bench of peers who are building the same thing.
Legal Transformation Institute · transformlegal.com
The forward deployed role is one of the fastest growing, best paid jobs in technology because it is what makes AI actually pay. These are the terms your firms and vendors already use. Here they are, defined plainly, so the standard is yours to set.