The 30-second version. A top medical device CLO just told me her CEO's mandate: most AI-forward company in the industry, and legal is not excused. Her team is building. Is yours? Somewhere in your organization sits a project everyone dreads, priced in the millions or never priced at all because the number was too out of reach. One person at one company just made a $4 million version of it vanish in overnight batches. My students just showed how to do it for a university's entire contract portfolio after a state changed its negligence law. Meanwhile only 17% of legal departments can measure what AI returns. This issue shows you how to find your AI impact, and how to tell whether anyone in your team can do the right thing with it.

TAKE THE DARN ASSESSMENT. Nine minutes to learn what skills you have and need to work on in terms of being able to build with AI. You get your own profile immediately, and your participation helps build a reference set that does not yet exist. [fdle.transformlegal.com/assessment]

When you get the results, review the following guide to understand what it means.

READ THE DARN GUIDE . I will not waste your time!. It is a comprehensive breakdown of the new role of legal engineer. This is for lawyers, leaders, and business professionals. It has real example of work and the depth you need to fully appreciate what skills are needed. Use this for yourself, to draft job descriptions, to inform interviews, and to assess your team. It is plain English, no AI background assumed, about twenty minutes: [fdle.transformlegal.com]

The $4 Million Stack

A story from the trenches. I literally just left a meeting with the chief legal officer of a major medical device company. I am typing this (yes me, not AI) as I dine by myself at a nearby restaurant.

Her CEO has declared the company will be the most AI-forward medical device company in the world, and legal is not excused. Why? Speed to market. A commercial goal, which is exactly where it needs to be. That means the in-house team is ramping. Palantir engaged. Foundry skills in the department. Claude rolled out at enterprise scale. They have a few non-technical builders, people who can build working tools but cannot enterprise-harden the final product. Many basic users. Some naysayers. And the CLO herself? A sophisticated non-technical builder.

Now, to the law firms reading this: how well will you show up when that client asks about your AI adoption? What do you have to share? Your spend? Seats activated, logins, attendance at the lunch-and-learn? Those are activity numbers. They measure motion, and her problem is speed.

No activity report closes that gap. Impact does. Problem is, nobody has found the work that make an impact to you or your clients yet. Up until recently, I have said that is okay as we are all on the same journey. But the more I meet with teams, dig into their work, and work along side them, I am revising this to – you need to show impact ASAP. I don’t care how – in your business model or for you clients. And the impact has to have commercial value.

Too many firms are shooting for big, juicy, and sexy proclamations of success. And they are missing. The value of this technology in a legal organization does not arrive as a massive disruption or innovation. It arrives in lumps. You need to fine the lumps. The lumps are findable once you know what to look for.

Two more stories

I keep hearing versions of the same story. In one of them, a major company was sitting on a document conversion project it had scoped and priced at $4 million. Thousands of documents, each needing expert review, each with a known per-unit cost as they had bid it out. One person looked at the stack, recognized it as a workflow a well-instructed Claude skill could tackle, and had a working version running within an hour! They built an eval next. And then batched all of the documents through it . . . still with human review but exponentially less. What had been budgeted at $4 million got done in overnight batches.

The second story I did not hear. I designed it.

Last semester my students worked with the Office of the General Counsel at Indiana University. The office manages hundreds of contract types across thousands of contracts. Some templated, many custom. Then Indiana changed a law on how negligence and gross negligence must be treated in contract language, with contingencies that touched the liability exposure of the university and everyone it contracts with.

Understanding the exposure sitting in those existing contracts meant a hand-to-hand review across the portfolio. Nobody ever priced that project. Everyone could see the quote would be outrageous, so the work got filed under impossible and left there.

My students built a prototype that did it for a defined slice of the portfolio. They mapped how the contract work actually flows, then did the legal work of defining, clause by clause, what counts as exposure under the new law. They made the tooling calls and ran the numbers on what the build was worth. A task the institution had written off as impossible became a scoped project with a method, and it took them one semester, alongside their other coursework.

What they built was a prototype, run against one defined slice of those thousands of contracts, and nothing has been deployed. Still, hold the two stories side by side. One stack carried a $4 million price tag and a lot of dread; the other had no price, because nobody wanted to see the number.

The fingerprint

So how do you spot this work? It looks the same everywhere I find it.

Start with this: the documents already exist. Nobody is writing anything new. You are taking what you already have and moving it from one shape to another. Old template to new template. Old law to new law. A drawer full of PDFs into a system you can actually search.

Second, you already know most of the rules. If you sat a smart associate down tomorrow, you could tell them how to handle ninety percent of it. The last ten percent is judgment calls, and those calls are the reason lawyers always had to do this work. Fine. Write them down. That is the whole trick, and it is the one part nobody ever bothers to do . . . because it is more fun to jump to AI.

Third, somebody put a price on it once. Go check your old vendor quotes. So much per contract, so much per file. And the biggest versions of this work never got quoted at all, because everyone could do the math. And that it why it gets ignored.

Fourth, you hate it. Be honest. It has been sitting in the backlog for two years because it is huge, it is boring, and whoever raises a hand to scope it inherits it. Queue Mark Twain’s frog on the side of the desk. (h/t to that CLO for this nugget).

Sound familiar? It should. If you lived through the GDPR repapering or the LIBOR amendments, you watched this industry pay nine figures for exactly this shape of work, priced by the contract. Maybe your version is the CLM migration you keep deferring. Or the contracts that still need to move onto the new templates from the merger. Or the engagement letters nobody has checked against the updated outside counsel guidelines. Or the same one my students caught: what does the new law do to every contract we have already signed?

Why nobody finds it

Notice what the tool was in both of those stories: software the organization already paid for. Nobody needed a new platform. They needed one person who could see the stack for what it was.

The secret sauce is that a person has to hold three things in their head at the same time. They have to know how the work actually moves, or they will never spot that the dreaded project. They have to know what the technology can do and what it cannot (accurately, predictively), or they build too much, or nothing. And they have to know the number. Not roughly. Well, enough to put it on one page that someone with a budget can say yes to so they can calculate the alpha (gains).

Miss any one of the three and the whole thing dies. I watch it happen constantly. The workflow veteran who does not know the tools? Files the project under impossible and moves on. The AI enthusiast who does not know the work? Builds a demo, shows it around, and nobody trusts it. And either of them, without the number, never even gets a meeting.

There is also a second half to both stories, and people leave it out every time they retell them. Why did the room believe the demo? Because the output matched what the experts sitting in that room already knew correct looked like. Somebody sat down beforehand, decided what correct meant, and checked. My students learn that as evidence discipline. Skip it and the same story becomes the cautionary kind. Finding the work gets you the meeting. Evidence gets you the deployment (and promotion? bonus?).

The stack audit

Okay. I have told you the stories. What the heck do you do with them?

Three things, starting tomorrow, and none of them require buying anything.

First, go pull your quotes. Three years back. Every project a vendor, an ALSP, or your own team ever priced per document, per contract, per file. Then make the second list, the one nobody wrote down: the projects that never got priced because everyone knew the number would be outrageous. Put the two lists side by side. That is your map.

Second, walk the map and ask two questions of every item. Does the content already exist? And is the job really a transformation, mostly rules with judgment at the edges? Circle every yes.

Third, and this is the one that matters: ask who in your shop can hold the work, the tools, and the number in one head. Got a name? Give that person a mandate or tee it up for them. No name? Then you just found your real AI gap, and it is a person gap, not a software gap.

The fourth question

The three skills in this letter are actually six. I have mapped the full set, the competencies of what I and others call the forward deployed legal engineer, in a guide you can read in one sitting. READ IT! Trust me – I will not waste your time. You will know more after!

Also, and this is critical, it ends with an interactive radar where you score yourself against all six.

Fair pushback: why score yourself at all? A skill set like this, you either have it or you don't.

No offense here but way too many firms and inhouse teams are misallocating AI talent. Right now, firms and departments pick their AI people by ear. The person who can drop the model names and hold a room with a demo: I call them the “AI entertained.” They are not charlatans – no shade. They simply do not know any better, and neither does anyone else in the room, because nobody can name what capability here actually consists of. So the mandate defaults to the entertaining, who chase the glamorous work, and talk in platitudes and AI jargon but have delivered zero alpha. Meanwhile the people capable of the unglamorous millions tend not to perform well in those rooms, so nobody thinks to ask them.

This is what the radar is for. Capability in this domain is illegible from the outside; it looks like nothing in a meeting. Scoring yourself makes the invisible skill set nameable, first to yourself, then in everyone you evaluate.

The have-it-or-don't framing also misreads what these skills are. Each of the six is a ladder of behaviors, and every rung is learnable, so a radar reads less like a verdict and more like a map with a next move on every axis.

Say you are the technical builder. You can make anything run, and your score confirms it, but you come up low on legal judgment and translation. There is the diagnosis of why your prototypes die in demos: you cannot yet say where a wrong answer creates real exposure, and you cannot yet pull what the expert knows into a spec anyone trusts. Read that way, your radar is telling you who to build with, and which two skills to borrow while you grow your own. This is a decoder for you!

Say instead you are the power user. A thousand chats deep, fast, and sure you could not live without it. Score honestly and the radar will show you something uncomfortable: fluency and capability are different things. You have never defined correct in advance. Nothing you have made runs without you. The gap between using and building is real, and the radar names the one or two artifacts that get you across it. This is a decoder for you!

And if you lead people: you cannot calibrate anyone against a skill set you cannot name. Score yourself so that the next time someone dazzles the room, you know the six questions to ask instead.

One warning. Expect the result to sting somewhere, probably on an axis you have not tried yet. When it does, sit with it. The score you want to argue with is the one telling you the most.

Two minutes. [FDLE guide link]

Nobody is born with this mix, and you cannot yet hire it off a shelf without making serious mistakes and paying dearly – trust me, everyone is trying and paying top dollar for it. It is trainable, though. I build students to acquire it in a semester and lawyers in less more intense time. The question worth sitting with is where the next generation of lawyers gets that mix at scale, because the reps that used to build this kind of judgment are the same reps AI is quietly removing.

That question is what this semester's experiment exists to answer. More on that as the data comes in.

Evidence, not vibes.

Josh

To read previous editions, click here.

DISCLAIMER: None of this is legal advice. This newsletter is strictly educational and is not legal advice or a solicitation to buy or sell any assets or to make any legal decisions. Please /be careful and do your own research.