Law firms are adopting AI for speed. In law, speed on unreliable reasoning is the most expensive kind.
The profession is racing to use AI for drafting, research and review. The hard part is not getting an answer faster. It is making sure the judgement underneath it is still yours.
Request a Strategic Briefing →Legal teams are reaching for AI to draft, research and review faster, and the fear underneath the excitement is what speed does to the billable model and to the firm's leverage. The trap is treating AI as a faster way to produce legal work. A fluent, confident, wrong answer in law is not a productivity gain. It is a liability event, and the courts have already sanctioned lawyers for relying on citations the machine invented. The advice, the risk position, the privilege call, what to rely on: none of that can be delegated to a tool built to agree with you. Gildoni installs the Havruta Methodology (formerly the Think Partner Methodology) into how lawyers reason with AI, so the firm gets the speed without surrendering the judgement its name is built on.
The whole profession is moving, and fast
This is not a fringe experiment any more. It is the direction of the whole market, and the people inside it know it.
The promise is obvious: AI drafts a clause, summarises a deposition, reviews a contract or surfaces authority in seconds, work that used to take a junior hours. The fear sits right beside the promise. If the machine does in minutes what the firm used to bill by the hour, the leverage model that funds the firm comes under pressure. So the adoption is happening at speed, often ahead of the policies and the training that would make it safe.
The regulators have noticed. England's High Court has been blunt that the risks of using AI for legal research are now well known, and that a lawyer who does not check the output has breached a professional duty. The promise of speed and the cost of an unchecked answer are arriving at the same time.
The numbers tell the same story: fast adoption, thin guard-rails.
Use of generative AI among legal professionals nearly doubled in a single year, and 95% expect it to be central to daily work within five years.
of legal professionals say their organisation has no policy on generative AI, and 64% have had no training in it. Adoption is outrunning governance.
cases cited in one High Court matter did not exist at all. The court set out sanctions running from costs and public admonishment to referral to the police.
The speed is real. So is the exposure.
Artificial intelligence is a tool that carries with it risks as well as opportunities. Those who use artificial intelligence to conduct legal research have a professional duty to check the accuracy of such research by reference to authoritative sources.
The misuse is treating AI as a faster typist
Here is where it goes wrong. The instinct is to point AI at the work and ask it to produce: draft the brief, find the authority, summarise the bundle. The machine obliges, fluently and at once. And that fluency is exactly the problem.
A wrong answer in law does not look wrong. It looks like a clean memo, a confident citation, a well-formed clause. The model is built to be plausible, not to be right, and it will produce a case that reads like real authority and does not exist. In most fields a plausible error is an inconvenience. In law it is the liability event: the negligent advice, the misstated position, the fake citation that draws a wasted-costs order and a regulatory referral.
Speed on reasoning you have not checked is not a saving. It is the most expensive thing in the building, because it moves an unverified answer into a client letter or a court filing faster than anyone can catch it. The lawyer who would never sign off a trainee's research unread will paste an AI answer into a pleading because it arrived sounding finished.
It is worth being honest about the deeper issue. The judgement at the heart of legal work, what the advice should be, what risk position to take, what is privileged, what to actually rely on, was never a typing task. It cannot be handed to a tool whose default behaviour is to agree with the way you framed the question. That is not a tooling gap. It is a reasoning gap, and no faster model closes it.
Getting the answer faster is the easy part
Every legal-AI vendor now competes on the same axis: faster drafting, faster research, faster review. Almost none of them change the one thing that decides whether the output is safe to rely on, which is the quality of the reasoning a lawyer brings to it.
That is the real gap, and it is not a speed gap. Put a question to a commodity AI and it hands back a confident answer without ever telling you what it is unsure of, what it invented, or what you failed to ask. At the altitude where the advice is given, that is the Mirror Principle at its most expensive: if the reasoning going in was thin, the legal position coming out is thin, however authoritative it reads. A faster tool does not make the thinking any better. It just delivers the unreliable answer sooner. The discipline that closes the gap is a different one.
Both paths start from the same question. The lower path runs straight to a fluent answer that arrives finished and lands as a liability event. The upper path adds the one step the speed race skips. The red is the distance between getting an answer fast and reaching a position you can actually stand behind.
What the Havruta Methodology installs in a legal team
The Havruta Methodology is that step. It changes the default behaviour of the machine from agreeing with you to reasoning with you, which is exactly what legal judgement requires.
The Flip
The Flip puts the machine on the other side of the matter. Instead of confirming the position, it argues against it: where is this authority weak, what has the other side got that you have missed, what would have to be true for this advice to be wrong. The reasoning gets tested before opposing counsel, or a regulator, does the testing.
Ground Truth
Ground Truth keeps the work anchored in the matter's real facts and verified authority, the actual file, the live case law, the client's real position, rather than the plausible-sounding average the model produces by default. It is the discipline that catches the citation that reads like law and does not exist.
Decision Velocity
And Decision Velocity is the honest version of the speed everyone wants. It compresses the path from question to a defensible position by improving the reasoning, not by skipping the check. The firm gets faster on the work it can stand behind, which is the only speed worth having.
The fuller account of how all of this works is on the methodology page.
What this is not
This is not legal-research software and it is not a contract-review tool. It is not e-discovery, not a drafting assistant, not a citation checker, and it is not AI training or general AI literacy. The tools are a separate market, and a useful one. This is the reasoning discipline that decides whether their output is safe to rely on.
It changes how lawyers reason about the work they already own: the advice, the risk position, the privilege call, what to rely on, and what to put before a court.
Where a firm starts
The methodology is installed along a ladder, and a firm or a legal team enters at the rung that fits.
Most begin with the Eye-Opener Workshop, a half-day in which the team sees the shift on its own real matters.
A practice group embeds the practice through the Havruta programme, taking the discipline across the team's daily work.
A single high-stakes question, the firm's AI risk position, a partnership-level call on the billable model, a governance decision, can be worked through Advisory Havruta.
How leaders reason with AI, by the role they hold
If you lead a firm or an in-house legal function, the role pages take the same discipline to the seat you sit in. A Strategic Briefing is how to decide where to begin.
Go to the leadership rolesWhat managing partners and general counsel ask
Is it safe to use AI for legal research and drafting?
It is safe when the lawyer treats the output as a draft to be reasoned through, not an answer to be trusted. The courts have been explicit that AI tools can produce confident, well-formed material that is simply false, including cases that do not exist, and that the lawyer who relies on it without checking has breached a professional duty. The risk is not the technology. It is using it as a faster typist instead of a thinking partner you interrogate. The discipline we install is exactly that interrogation.
Will AI undercut the billable-hour and leverage model?
It puts the model under real pressure, and pretending otherwise helps nobody. If AI does in minutes what a junior used to bill for hours, the leverage that funds the firm shifts. The honest response is to decide deliberately how the firm captures value when speed is commoditised, rather than letting individual lawyers improvise. That is a partnership-level reasoning question, not a tool rollout, and it is precisely the kind of high-stakes call the methodology is built to work through.
What is the real risk of AI hallucination in legal work?
The risk is that the error is invisible. A hallucinated citation does not look wrong. It reads like real authority, formatted correctly, and slips into a memo or a filing because it arrived sounding finished. In one High Court matter, 18 of 45 cited cases did not exist, and the court set out sanctions running from costs to regulatory referral. The lesson is that a fluent wrong answer in law is a liability event, so verification against an authoritative source is not optional.
Is this legal-tech software or a contract-review tool?
No. It is not legal-research software, not a contract-review or e-discovery platform, not a drafting assistant, and it does not touch your stack. Those tools speed up production. This addresses the thinking around them: how a lawyer reasons through an AI-assisted answer so the advice is truly theirs, anchored in the real file and verified authority, and stress-tested before opposing counsel or a regulator does it for them. The tools and this discipline are complementary, not competing.
How should a firm govern AI use without slowing everyone down?
Stop treating governance as a policy document and start treating it as a reasoning habit. A policy nobody internalises does not survive contact with a deadline. The better answer is to install the discipline directly into how lawyers work with AI: make the machine argue the position rather than confirm it, anchor it in verified facts and authority, and verify before reliance. Done this way, the speed comes from sharper reasoning, not from skipping the check, which is the only speed a firm can defend.
Where should we start?
With a Strategic Briefing, or with the Eye-Opener Workshop, where a legal team sees the difference between instructing AI and reasoning with it on its own real matters. From there the path depends on whether you are setting the firm's AI position, embedding the practice across a practice group, or working a single high-stakes question such as the future of the leverage model.