Memorandum není výstup

A general counsel asks their team a question. Can we end this distributor agreement in the Gulf without triggering a compensation claim? Two days later a document comes back. It is careful, well researched and correct. It sets out the governing law, the relevant clauses, three scenarios and the risk attached to each.

It does not say what to do.

The business reads it and asks the only question they had in the first place. So what do we do? A capable lawyer has just done two days of real work and produced something the organisation cannot act on. Nobody in this story is bad at their job, and everybody in it is frustrated.

The seat moved. The deliverable did not.

The Association of Corporate Counsel runs an annual survey of chief legal officers. For the 2026 edition, its twenty-seventh, it surveyed 1,049 chief legal officers in 43 countries. It found that 84% now report to their chief executive, up from 79% the year before, and that 52% keep a direct line to the board.

That is a structural change, and it is worth sitting with. The senior lawyer is no longer the person who checks a deal after it has been designed. They are in the room while it is being designed, which means they are being asked commercial questions in commercial language, on commercial timelines, by people who will act on the answer.

The output that legal was trained to produce was built for a different job. It was built to be defensible, complete and reviewable. Those are the right qualities for an opinion that may be read years later by a court. They are the wrong qualities for a decision that has to be made on Thursday.

Why the memo persists

This is a training artefact rather than a character flaw, and it is worth being precise about where it comes from.

In private practice, the comprehensive risk analysis is frequently the correct deliverable. An external firm is retained to give the client the full picture so the client can make their own commercial decision. Handing over a recommendation the client did not ask for can be presumptuous, and occasionally it is negligent. Lawyers are taught this early and they are taught it for good reasons.

The habit then travels in-house, where the economics are inverted. Nobody in the company is going to take the picture and turn it into a decision, because the person who was supposed to do that is the lawyer. There is also a quieter incentive at work. A document that lists every risk is very hard to be wrong about. A document that says do this is falsifiable by Friday.

A memo hands the business the whole keyring. A recommendation hands them the one key that opens the door, and takes responsibility for the choice.

The economics stopped being patient

For a long time this was a matter of style. It is now a budget question.

The Corporate Legal Operations Consortium published its 2026 State of the Industry report in March, drawn from the 135 legal departments it surveyed. Only 37% expect outside counsel spend to increase, down from 58% the previous year. Only 47% expect their internal legal budget to grow, down from 65%. Barely a third expect to grow their legal headcount.

Meanwhile the cost of buying the work outside went up. The Thomson Reuters Institute and Georgetown Law reported in January that worked rates at US law firms grew 7.3%, a record, with some firms pushing annual increases of 10 to 15%. That figure is drawn from the US market rather than a global one, but the direction is the pressure every general counsel is now managing.

Flat headcount, flat budget and rising external rates produce one conclusion. Every hour spent producing analysis that nobody acts on has to come out of somewhere, and it comes out of the work that would have moved something.

What actually changes in the document

The shift is smaller than people expect. It is not a different skill set, and the underlying legal work is identical. What changes is the order, the editing and who carries the ambiguity.

The right-hand column is harder to write. It requires the lawyer to decide which of the risks they found are actually load-bearing for this decision, and to be visibly wrong if they choose badly. That discomfort is the job now.

Two things make it survivable. The first is stating a confidence level out loud, because a recommendation that names its own reservations is more useful than false certainty, and far more useful than silence. The second is naming what would change your mind, which converts a decision into something the business can monitor rather than something they have to trust.

Where AI helps, and where it quietly makes this worse

Wolters Kluwer surveyed 810 legal professionals for its 2026 Future Ready Lawyer report, across the US, China and nine European countries. It found that 92% now use at least one AI tool daily. Inside corporate legal departments the split is the interesting part. 56% reach for general-purpose generative AI, the same assistants everyone else in the company uses, while only 14% use tools built for legal work, and 40% report using none at all. The survey does not cover the Middle East, so read it as a signal about direction rather than a regional measurement.

That 56% is where the risk sits, and not for the reason people usually give.

Fluent, well structured, comprehensive-looking analysis is the single easiest thing for a general-purpose model to produce. So the obvious first use of AI in a legal department is to generate more of the exact deliverable that was already the problem, faster and at greater length. The memo was a bottleneck. Automating it does not remove the bottleneck, it industrialises it.

The useful direction is the opposite one. Compress the research, the first-pass review and the document gathering, so that the lawyer still has judgement left at the end of the day to form and defend a view. The recommendation is the part that does not automate, and it is the part the business is actually paying for.

There is a second problem with reaching for a general assistant here. A recommendation you sign your name to has to be traceable back to something. Which contract clause, which statute, which version of the regulation, in which jurisdiction, as it stood on which date. A model that is fluent everywhere and grounded nowhere in particular will produce something that reads exactly like an answer, and you will not be able to check it in the ten minutes before the meeting.

What we build for this

This is the problem HAQQ is built around, so we should be direct about how, and about what that does and does not solve.

Justinian®, our legal reasoning engine, is designed so that an answer arrives with its sources attached rather than as a paragraph of confident prose. The intent is narrow and deliberate. A lawyer should be able to see what a conclusion rests on, open the underlying provision, and decide whether they agree, without leaving the answer and starting the research again from the beginning.

Underneath, a question is routed rather than passed straight to a model. What kind of legal work is being asked for, which jurisdiction governs it, what sources are authoritative for it, and what has to be retrieved before anything is drafted. We wrote about that architecture separately, because it is the part of legal AI that is invisible from the outside and it is the part that determines whether the citations hold.

What this does not do is form the view. It gives back the hours that used to go into assembling the picture, and it makes the picture checkable. Deciding what the company should do, and putting your name on it, stays where it has always been.

Four questions before you send it

• Does the first line answer the question? Not the background, not the governing law. If a busy executive reads only the opening sentence, do they know what you are recommending?
• Have you said how confident you are? A number or a plain sentence. Unstated uncertainty gets read as certainty, and that is how legal advice gets misused.
• Have you said what would change your mind? One or two facts that, if they turned out differently, would flip the recommendation. This is what lets the business act now and stay alert.
• Have you asked for something specific? A decision, an approval, a document, a date. Please revert with your instructions is not an ask, it is a way of handing the problem back.

None of this makes the legal analysis lighter. The research still has to be right, and being wrong in a recommendation costs more than being vague in a memo. That is the trade, and it is the one the seat at the table came with.

• The Justinian engine, and how it reasons
• What happens before a legal AI answers your question
• How a law firm becomes AI-native
• 45 red flags when evaluating a legal AI vendor