I asked an attorney at my firm to have the AI draft the initial pleadings on a live matter. It thought for a long stretch and then refused. It was right, and that refusal was the most useful thing that happened in my office that week. AI legal drafting only breaks in the direction of your inputs, and this one broke honestly.

Two blockers, and both of them were accurate

The refusal was specific rather than vague, which is the part worth paying attention to. It named two separate problems. The matter was not covered by any signed agreement, and evidence gathering was not finished either. The full representation engagement agreement was unsigned. So was the petition. On the intake side, items were still outstanding.

Two separate problems. It isn’t covered by any signed agreement. Evidence gathering isn’t done either.

I confirmed on the record that every blocker was accurate. There were no completed intake forms on that matter. There had never been an intake call. Nothing the tool reported was invented and nothing was excess caution. It read the file, compared what was there against what drafting a petition actually requires, and reported the gap without softening it.

The override failed, because AI legal drafting cannot invent inputs

I tried to push through it. Search the file and proceed with drafting from what you know. That produced no draft either. There is no prompt that manufactures a signed engagement agreement, and no phrasing that conjures an intake call that never happened. The tool can only be as good as what you feed it and what you correct. A firm that expects AI legal drafting to paper over a missing foundation is asking software to commit the firm to something the firm never actually did.

A broken law firm intake process feels exactly like broken software

This is where firms decide the product is defective. It is not defective. It is auditing the law firm intake process, out loud, in front of whoever happens to be on the call. Every refusal reads as an obstacle when what is really happening is that a system with no social skills is repeating the thing your staff already suspected and never escalated. Most of the frustration I hear about AI for law firms is exactly this, described from the wrong end.

In this instance, a human at the firm had asked out loud, before the AI said anything, whether we were even engaged on that matter. The question got asked and dropped. The AI asked it again and would not move on. That is the whole difference between a person raising a concern and a gate enforcing one.

There is a second-order point in the same fact pattern that has nothing to do with technology. The client had more than one matter. A signature on one is not coverage for the other. Scope runs per matter, and the moment a firm starts drafting outside the scope it was retained for, it has a billing problem and a professional responsibility problem before it has an AI problem. The AI happens to check that every single time, unprompted, without getting tired at five o’clock on a Friday.

Intake is a gate, not a formality

The rule I run in my own office is the one I give attorneys who ask me why their systems keep stalling, and I state it the same way every time.

If you go into your doctor’s office and you have not completed the intake forms, you are not going to see the doctor. In my office, that’s how this works.

Operationally that means drafting does not start until a signed engagement agreement covering that specific matter is in the file, the intake form is complete, and the intake call has happened. Not two out of the three. The advantage of putting that gate inside the system rather than on a whiteboard is that the system does not make exceptions for a client it likes, does not forget on the day the office is short-staffed, and does not quietly assume somebody else already handled it. It also cannot be talked out of it, which is inconvenient once or twice a year and worth it every other week.

Two other things broke that afternoon, and both were mine

Honest texture from the same session, because the useful version of this story is not the flattering one. I could not open a resource I had built myself, because I had granted access to everyone else and never to myself. My entire diagnosis was that I did not give myself access. Then a configuration audit could not establish what was actually installed on the other machine, and I ended up saying that I do not understand what I do not know. Both of those are permission and inventory problems, not intelligence problems. They are also exactly the kind of thing that makes an attorney conclude the technology is unreliable, when the real issue is that nobody owns the setup. Who can see what inside your systems deserves its own decision, which is why I treated attorneys, AI and client confidentiality on its own rather than as a footnote.

When the configuration is right, what you are working with is closer to a capable new associate than to a machine. You are not getting a kindergartener. You are getting a high schooler. It has judgment, it will tell you no, and when it tells you no it deserves ten minutes of your attention before you overrule it. A research tool that refuses to invent authority is the same feature wearing different clothes, which is the argument in the research library that refuses to do research.

Run this audit on your last ten matters

Pull the last ten matters your firm opened. For each one, ask whether there is a signed engagement agreement in the file naming that specific matter, a completed intake form, and a documented intake call. Whatever fraction fails is exactly the fraction where AI legal drafting will stall on you, and it is also your exposure with or without any software. The AI is not the reason to fix intake. It is just the fastest audit of intake you will ever run, and it is free. If the folders themselves are disorganized you will get false failures on top of the real ones, which is a separate and very common problem I covered in why your AI did not fail and your file structure did.

None of this is legal advice or a compliance opinion. Every attorney is responsible for their own compliance with their bar’s rules on engagement, scope and the use of these tools.

If you want a system that enforces your intake gate instead of arguing with it, book a custom build strategy session at calendly.com/leisawintz/lal-custom-ai-strategy and bring your current intake workflow, however ugly it is. We will map where it leaks before we talk about building anything on top of it.