The 30-second version. Deploying AI inside legal work takes six capabilities. Five of them are not new; they have simply been renamed. You may already hold three or four without recognizing them. One is genuinely new to most legal professionals, and it is where nearly all the certification energy is going. That mismatch is one reason deployments struggle. I built two things to help you see the whole picture. Open them but come back to read this.

The field guide. A comprehensive breakdown of the role with 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 assessment. Nine minutes. You get your own profile immediately, and your participation helps build a reference set that does not yet exist: [fdle.transformlegal.com/assessment]

Law firms are beginning to hire, train, and organize around a role they have not clearly defined.

It is called the forward deployed engineer, the forward deployed legal engineer, the legal engineer, or some variation of those terms. The title changes. The underlying expectation does not: find someone who can move artificial intelligence from an impressive demonstration into real legal work.

That person is now appearing in job descriptions, team designs, compensation discussions, and certification programs.

But the role is being defined too narrowly.

The market is concentrating on the most visible part of the job: technical fluency. Can this person use the model, configure the platform, build an agent, or connect the tools?

Those capabilities matter. But they are only one part of the work.

A successful legal AI deployment also requires someone—or, more realistically, a team—to understand how the work actually moves, where legal judgment must intervene, how expert knowledge becomes a buildable specification, how the system will be tested, and whether it makes economic sense.

That creates a practical problem for law firm leaders.

You can write a job description for the wrong person. You can hire technical talent into a system that is not prepared to use it. You can send people through certification programs without knowing which capability the team actually lacks. And you can overlook experienced people already inside the organization because the market has given unfamiliar names to skills they have spent years developing.

This issue is my attempt to make the role legible.

I have broken the work into six capabilities. Five are established disciplines operating under newer names. One is genuinely new to most legal professionals. Together, they provide a more useful way to assess individuals, assemble teams, write job descriptions, evaluate candidates, and understand why promising AI deployments stall.

If you lead a firm or legal department, this is a framework for deciding what capability you actually need.

If you are building one of these teams, it is a way to see what you already have before hiring what you do not.

And if you are wondering where you fit in this emerging field, you may be further along than you think.

---

THE FULL BRIEF

We are all on the same path, just at different points

Everyone is working through this. The field is moving too quickly for any of us to see all of it, and confidence is not always a reliable measure of exposure.

I have been working on the industrialization of legal work for twenty-five years, and I am still learning something new every week.

That is the honest state of the field. It should be a relief, not a worry.

But part of what is happening is doing real damage, and it is not the technology.

The market renames things constantly. Every month brings a new word for an idea that has been around for twenty years. The new word arrives with a conference, a job posting, a new AI expert on LinkedIn, and a price tag attached. The latest is the “legal engineer.”

That does three things to you. All are bad.

First, it makes you feel behind on capabilities you already possess. If you have spent a decade mapping how work actually moves through a legal department, you hold a capability that is now being renamed and sold back to you as a scarce new skill. You did not lose it. Somebody renamed it while you were busy doing it.

Second, it makes you over-index on whatever has the newest name. Because the new term describes the thing you do not have, it feels like the gap. It may be a gap. It does not follow that it is your most important one.

Three, it obscures what the true nature of the work actually is. There might be a prior understanding of the work, but the new label blurs what people are referring to and makes the old label feel antiquated, even if the skills it represents are as needed and current as ever.

This is why I decided to decode the forward deployed engineer role once again, but this time I am going deeper. I have had a handful of conversations with people in similar roles. I have also examined my own work.

Why this role? Because it has become the zeitgeist catch-all for whatever legal teams supposedly do not have, suddenly need, and can now acquire through a credential.

I do have a warning for you. And like I already shared above, I built two things for you, because building things is what I do when I do not know what else to do with a problem.

Five of the six are not new

Through my recent work building with AI for myself and clients, and my years building new business models inside legal teams, I have worked through what successful deployment actually requires.

I have landed on six capabilities. Together, I call them the Hexagon. Five are established disciplines operating under newer names.

Workflow intelligence is process engineering coupled with dose of Elon’s algorithm. It means decoding how work really happens and should happen, task by task, with its exceptions and routing logic included. Full disclosure: I am using “workflow intelligence” here, which makes me guilty of exactly what I just warned you about above – old skill, new name. My partner Rob Saccone is a Six Sigma Black Belt. I am a Green Belt. This discipline is older than either of us. Our colleague Kim Craig was a pioneer in bringing it into legal over two decades ago. These are credentials that demand you prove your work and impact before you get them, rather than attending and getting a new badge.

Legal judgment is what lawyers begin developing in law school and refine through practice: knowing what defensible work looks like and where a wrong answer creates real exposure.

Commercial judgment is firm economics. Rates. Realization. What something costs per run and what it returns. Whether it is worth building at all. Partners who have built books of business hold some of this. Some pricing professionals hold it deeply; others have never been given the authority or operating information required to develop it. Right now, many firms are building with AI and punting on the coming economic and commercial realties.

Translation is business analysis and communication. It means getting what an expert knows but cannot articulate out of that person's head and into a form a builder can use, then explaining and defending the finished thing back to the expert. Product managers and business analysts have been doing versions of this for decades.

Evidence discipline is quality assurance: test cases with the answers decided in advance, a pass rate, failure categories, and a rule for what must reach a human. It is old in software and still unfamiliar in much of legal. That is a different problem from being new to the world.

Technical competency is the genuinely new component for most legal professionals. Models, agents, guardrails, integrations, and knowing when AI is overkill. It is real. It is hard. Nobody gets to skip it.

One of six is new.

Read the list again and count how many you may already hold to some degree.

A litigation support manager who ran review workflows for eight years holds one outright and probably two. A senior associate has spent years developing legal judgment. A pricing director may already possess commercial judgment. A legal operations professional may have deep workflow intelligence and translation ability without ever using either label.

Most of you are further along than the discourse has led you to believe. You have been quietly discounting yourself because you cannot yet speak the vocabulary. And yes, many of you are not as far along as you think. Hubris reigns supreme in many pockets of firms, and no team can outrun Dunning-Kruger.

The vocabulary is the easy part. It is a glossary. It takes an afternoon. Which brings me to my warning.

Now the warning

Two warnings, actually.

You are probably over-indexing on technical fluency. It is the newest capability, the most visible, the most certifiable, and the easiest one to feel behind on. It is also only one of six. Chasing it first, without examining the capabilities you already hold and the problem you actually need to solve, is a common mistake.

And you may not have as much technical fluency as you think.

Using AI tools every day gets you technical fluent. Being good at prompting is technical fluency.

Technical competency is being able to look at a task and determine which parts belong to fixed-rule software, which belong to a model, and which must remain with a human; and then defend that division. Many daily AI users cannot yet do that. There is no shame in it. But the confusion is expensive.

Both things can be true at once: you are further along than you think on some axes and further behind than you think on another.

That combination is the actual situation for almost everybody. It is also why a single certificate cannot tell you where you stand.

I have seen this movie

I have watched a new category of legal work emerge before.

I entered electronic discovery before it had settled into a recognized discipline. By the middle of the 2000s, I was building and running document reviews for BigLaw clients, sometimes with as many as three hundred contract attorneys working at once.

As the industry matured, software platforms introduced certifications. The credentials were useful: they showed that someone understood a particular tool and could operate it competently.

But they did not tell me who could run the work.

The people I promoted were the ones who could see the entire system. They understood what was being routed where, which judgment calls were being made by the wrong person, where the review was producing confident garbage, and what each additional pass cost relative to the risk it reduced.

They could translate between the lawyers, the technology, the workflow, and the economics. When something went wrong, they could locate the actual cause rather than simply operate the software more aggressively.

Some were certified. Some were not. Certification was one useful signal, but it was not the same thing as effectiveness.

That is the pattern I see returning now.

AI certifications can establish knowledge of a model, platform, or defined technical practice. That knowledge matters. But deploying AI inside consequential legal work requires more than operating the technology.

The credential may tell you whether someone understands the tool. It cannot, by itself, tell you whether that person can see the work.

Which brings us back to this month

Anthropic previously introduced a technical certification for architects in its partner ecosystem. Harvey Academy now offers certifications in foundational legal AI and applied legal workflows.

These are real. They can be useful. If someone on my team wanted to pursue one, I would ask a simple question: Why?

How will this certification make you more successful in your particular job, team, and environment? It will make you more capable in the body of knowledge it covers. But is that the capability your team needs most?

Building in a vacuum is a hell of lot easier than building in an existing organization.

That is not a criticism of either program. A vendor certification is, correctly and by design, bounded. It validates knowledge associated with that provider, its technology, or a defined use of it. The providers are doing exactly what they should be doing.

But a certification validates a slice of capability. It cannot tell you whether that slice is the one your context requires next.

That is the larger issue. Most certification energy is flowing toward the newest and most visible component. Far less attention is going to the five older capabilities that are harder to see and often determine whether a deployment survives contact with real legal work.

A widely cited 2025 MIT NANDA preprint estimated that 95 percent of the enterprise generative AI initiatives it examined showed no measurable profit-and-loss impact. That is one study, not a universal failure rate, and “no measurable P&L impact” is not the same as saying the technology failed.

The study does not establish why every individual initiative stalled. But its findings are consistent with what I repeatedly see in legal deployments: the model is rarely the entire problem.

Getting the model to work is the part that got easier. Deployments struggle when the surrounding disciplines are assumed, overlooked, or undervalued.

➡ Nobody mapped how the work actually happens and should happen, so the system automated a fiction.

➡ Nobody located where a wrong answer creates exposure, so there was no escalation rule worth the name.

➡ Nobody wrote a specification, so the builder guessed.

➡ Nobody built an evaluation, so there was no way to distinguish a system that works from one that merely sounds like it works.

➡ Nobody ran the numbers, so nobody could say what it returned.

Six capabilities. Most credentials concentrate on one. The deployment still depends on all six.

What I am not going to sell you

I am not going to certify you. Not now, not later.

I do not hold the Anthropic certification either. I considered it. What I did instead was spend three years teaching myself the two components I did not have while strengthening the others, then prove it the only way that ultimately counts: by building an agentic system now running inside one of the fifty largest firms in the country.

That is a better path. It is a slower one. It was the path available to me.

That is the point.

Your context determines what you need next. In-house or firm. Lawyer, technologist, operator, or leader. The problem you are trying to solve. The capabilities your team already covers without anyone having named them.

A general counsel with a three-person department and an AmLaw 50 practice group need different things. Neither necessarily needs all six capabilities inside one human being.

What is common is the six. The path through them is yours.

As I said, I built you something

Two things went live this week.

A field guide. This is a deep dive into the work of forward deployed legal engineering. There is real work in it. It is written in plain English, with no AI background assumed. It explains the six capabilities, how the work actually runs, what a defensible system looks like inside a firm, and a nineteen-word glossary that does the decoding I described above. It takes about twenty minutes. It is free, and nothing is gated. Read the field guide.

A self-assessment. It takes about nine minutes. You answer twenty-four questions across the six capabilities and receive your own profile the moment you finish. It is not a score against a bar. It is a shape: what you already hold, where you are thin, and where your confidence may be running ahead of or behind your evidence. Take the assessment.

This is version 1.0 of a self-assessment, not a validated instrument. The initial benchmark and instrument study are being conducted with Indiana University Maurer School of Law and Ice Miller LLP.

The ask, and a condition

I need at least 250 completed assessments before the aggregate results can tell us much beyond each participant's individual profile.

At 250 completed assessments, I will publish the first composite view of where respondents sit across all six capabilities. I will also publish comparisons by role and organization type wherever the response counts support a credible comparison.

That reference set does not exist today. It will begin to exist when enough of the right people answer twenty-four questions honestly.

So please share it. Send it to one person on your team, in your department, or at your firm. There is no shortage of people interested in this field. Finding 250 willing to examine their own experience honestly should be achievable. I would like to go well past it.

Now the condition, and I mean this.

Be honest.

The assessment asks what you have actually done, not how good you feel. There is no leaderboard, no score to beat, nobody watching, and no title waiting at the end. If you inflate your answers, you will get back the profile of a person who does not exist. The only person you fooled will be the one reading it.

The assessment also measures confidence separately from evidence, on purpose. If your confidence outruns what you have actually done, the instrument will show you the gap. If your evidence is stronger than your confidence, it will show you that too.

Honest lows are more useful than generous middles.

---

Go get the certificate if it serves your purpose. It is real. It may be useful. And it covers only part of the picture.

Then look at the other five.

Most of you are going to be surprised, in both directions.

What do you already hold that nobody ever thought to name?

Talk soon again,

Josh

AI tool sprawl happened fast, and visibility never caught up. Employees paste contracts into ChatGPT, run code through Copilot, and build workflows in apps security never approved.

Harmonic Security classifies every AI interaction by task, tool, and team, so you can see which use cases drive real productivity, which tools are shelfware, and where sensitive data is headed. Across approved and unapproved apps alike.

No need to guess what 'AI adoption' means inside your company.

Explore Your AI Usage

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.