How to Put Your Firm's Experience to Work with AI

A first draft arrives. It is fluent, neatly structured and recognisably an agreement.

Then the partner starts editing.

The definitions are not how the firm writes them. The liability clause takes the wrong starting position. The termination provisions miss a protection the team usually includes. Much of the document is usable, but it does not feel like the work of an associate who knows the firm.

The problem is not simply whether AI can draft. It is whether it has been given the context that makes a draft useful.

Your precedents contain more than clauses

Over years of practice, a firm develops its own way of working. Partners favour certain structures. Teams learn which provisions need particular care. Different clients have different negotiating positions and expectations.

Some of that knowledge appears in formal templates. Much of it sits in completed agreements, marked-up drafts and the correspondence explaining why a clause changed.

Taken together, these materials hold something more valuable than a collection of reusable words. They record how the firm approaches its work.

Call it the firm's legal DNA.

Not every sentence in an old agreement deserves to be repeated. But the reasoning behind the firm's choices is worth preserving and making easier for the next person to find.

A good associate does not start with a blank page

When you ask an associate to prepare an agreement, you rarely give them only the agreement type.

You explain who the client is, which side you act for and what matters commercially. You point them towards a suitable precedent. You may add: "Use this structure, but not the indemnity from that deal. That was a concession."

Those instructions are the difference between producing a document and preparing a draft that moves the matter forward.

AI needs that distinction too.

"Draft a services agreement" leaves important choices unspecified. A more useful starting point identifies the client's position, the relevant precedent, the agreed commercial terms and any provisions that require a fresh approach.

The objective is not to make every document sound the same. It is to carry forward the right choices without carrying forward the wrong assumptions.

The final signed agreement is not always the best starting point

Imagine a firm acting for a supplier on a new services contract.

Its most recent signed agreement may look like an obvious precedent. But that agreement might contain a broad indemnity accepted for a particular customer, or an unusual termination right negotiated in exchange for a longer commitment.

Copying those provisions into the next draft could turn a one-off concession into the firm's default position.

A useful drafting process therefore needs more than access to past documents. It needs a way to distinguish an approved starting position from a negotiated outcome.

Which document should guide the draft? What was specific to the earlier matter? Which clauses are still suitable? Where does the lawyer need to decide again?

Finding an old agreement is only the beginning.

Start with the right examples, not the entire archive

Making better use of a firm's knowledge does not have to begin with every file it has ever produced.

A more practical starting point is one recurring agreement type and a small, carefully selected set of materials: an approved precedent, drafting notes and examples that explain important variations.

Record the preferences that otherwise have to be repeated. How should the document be organised? Which terminology should it use? Which clauses are preferred starting positions? What must be flagged rather than assumed?

Keep the material current, limit access appropriately and respect confidentiality and any restrictions on reuse. A document from another client's matter is not automatically suitable simply because it is available internally.

Then test the result on a realistic task. Does the draft follow the instructions? Does it use the correct starting position? Can the reviewing lawyer identify the material it relied on and the questions it could not resolve?

That is a more meaningful test than whether the prose sounds convincing.

Legal DNA should mean continuity, not blind imitation

An AI associate should not be judged only by how quickly it fills a page.

The more useful question is whether it helps the lawyer build on the firm's knowledge without losing sight of the current matter.

That means retaining familiar drafting conventions where they help, recognising when an old provision does not fit and making uncertainty visible instead of quietly filling the gap.

It also means being precise about what "learning the firm's style" involves. Providing selected precedents and reusable instructions is not the same as automatically training a model on the firm's entire history. Firms should understand which approach a product uses and how their information is handled.

Twenty years of experience should not have to be explained from scratch with every new draft. Nor should twenty years of documents be copied without question.

The opportunity is to make that experience a better starting point — while leaving the decisions that matter with the lawyers responsible for the work.