AI for Law Firms: Where It Belongs and Where It Absolutely Doesn't

We have not built a system for a law firm. We are writing this because we are asked about it regularly, and because the answer is more constrained than it is in other sectors — which is exactly why it is worth setting out honestly rather than pitching.

This is not legal advice, and it is not a claim of legal expertise. It is an engineering view of which parts of a firm's work have the shape that automates well, with the professional obligations flagged where we know they apply.

The three places it genuinely fits

1. Intake

The clearest case, and the one most firms feel first.

A prospective client rings or emails. Someone takes their details, works out what area of law it falls under, checks for conflicts, and decides whether it is worth a consultation. Most of that is structured questioning.

What an agent can do: answer after hours, capture the details, ask the standard qualifying questions for that practice area, run an initial conflicts check against your existing matters, and book a consultation.

What it must not do: tell them whether they have a case. That is advice.

The commercial argument is speed. Someone who has just been dismissed, or served, or had an accident, rings several firms. The first one that answers and books them in has an enormous advantage, and legal enquiries cluster heavily outside business hours because that is when people have time to deal with their problem.

2. Document assembly

Firms produce a large volume of documents that are structurally identical and factually different. Engagement letters, standard agreements, court forms, correspondence following a known pattern.

What automates: assembling a draft from matter data — the parties, dates, amounts, jurisdiction — into your firm's precedent.

What does not: drafting anything where the wording is the work. A clause negotiated to allocate a specific risk is not a template exercise.

The distinction is whether you are filling in a precedent or exercising drafting judgement. Firms usually know which is which instinctively.

3. Matter administration

The unglamorous one, and often the biggest time saver.

Deadline tracking and limitation periods. Filing documents against the right matter with consistent naming. Extracting key dates from incoming correspondence. Chasing outstanding client information. Time-recording prompts.

This is mechanical, high-volume, and it is what junior staff and paralegals lose hours to. It is also where a missed date is catastrophic, so the checking layer matters as much as the automation.

Where it does not belong

Anything constituting legal advice. An agent must not tell a client whether they have a claim, what their prospects are, or what they should do. Beyond the professional conduct problem, it is the single fastest way to a complaint you cannot defend.

Anything you sign. If your name is on it, you reasoned through it. A system producing output a solicitor rubber-stamps is worse than no system, because it manufactures confidence without the underlying work.

Advocacy and negotiation. Not automatable, and not the part anyone wants automated.

Case outcome prediction. Products exist. Treat them with heavy scepticism, particularly in Australia where the data is thin compared to jurisdictions these tools are usually trained on.

The obligations that shape any build

We would want your input on all of these before designing anything, because they are constraints rather than preferences.

Client confidentiality and legal professional privilege. Where does the data sit, who can access it, and does processing offshore create exposure? For many firms this alone rules out consumer-grade tooling.

Conflicts. Any intake system has to check against existing matters properly, and the failure mode of a missed conflict is severe.

Trust accounting. Regulated and audited. Not a place to experiment.

Supervision obligations. Work produced under a practitioner's supervision remains their responsibility regardless of what drafted it.

Record-keeping and retention. Often longer than you would otherwise keep data, and set by the obligation rather than by preference.

Costs disclosure. If an intake agent discusses fees at all, it must not stray into something that could be characterised as a costs agreement.

What we would build first, if asked

If a firm came to us tomorrow, we would almost certainly start with after-hours intake — not because it is the largest saving, but because it is the lowest risk and the fastest to demonstrate.

It sits entirely before the solicitor-client relationship begins. It gives clear commercial signal within weeks. And it establishes whether the firm is comfortable with how the technology behaves before anything touches live matters.

Document assembly second. Matter administration third, because it needs the most integration with systems you already run.

What we would not do is start with anything client-facing beyond intake. The risk profile is wrong, and a firm that has not seen the technology work in a low-stakes context has no basis for trusting it in a high-stakes one.

An honest position

We have built agentic systems handling document processing, financial verification and after-hours contact in other regulated sectors, including health and finance, where the privacy and accuracy constraints are comparably strict. The engineering transfers. The professional obligations do not, and they would need to come from you.

If you run a firm and are considering this, the useful first conversation is not about technology. It is about which of your recurring processes are genuinely mechanical, and where your professional obligations draw a line that no system should cross.

We would rather have that conversation and conclude that the answer is "not yet" than build something a practitioner cannot comfortably put their name near.