AI in legal practice — what it actually changes

Where we are now

Legal AI has moved from proof of concept to practical use in the space of two years. Tools exist today for contract review, legal research, drafting support, client intake, matter summarisation and workflow automation. Some of these tools are good. Some are not.

The firms and in-house teams getting real value from AI are not the ones with the most technology. They are the ones who identified a specific problem first, chose the right tool second, and built the habit last.

That order matters. AI bought without a use case becomes shelf-ware. A use case pursued without the right tool creates frustration. Neither is a technology problem — both are strategy problems.

What it actually changes

The honest answer is that AI does not change legal judgement. It changes the time cost and cognitive load of getting to that judgement.

A lawyer reviewing a 200-page contract still needs to understand the commercial context, spot the issues and advise the client. AI can accelerate the review, flag deviations from standard positions, and produce a first-pass summary — but the judgement remains human.

That is not a limitation. It is the correct frame. The tasks AI handles well are:

1. Document review and comparison. Reviewing contracts against a playbook, comparing versions, flagging non-standard clauses. This is well-suited to current AI tools and the time saving is significant at volume.
2. Research and summarisation. Pulling relevant cases, summarising legislation, producing first-draft research notes. Still requires verification — but the starting point is materially better and faster.
3. Drafting support. First drafts of standard documents: NDAs, employment contracts, board minutes. AI produces a serviceable starting point; the lawyer adds the judgement and specificity.
4. Client-facing intake. AI agents can handle initial client queries, triage matters, capture information and route to the right fee earner — reducing admin load without reducing service quality.
5. Matter summarisation. Long matters with extensive correspondence can be summarised for handover, supervision review or client reporting. Time consuming when done manually; fast with AI.

What AI cannot do

There is a lot of noise about what AI will eventually do. For practical planning purposes, what matters is what it cannot reliably do now:

1. Apply judgement in novel situations. Where there is no clear precedent or the facts are genuinely ambiguous, AI produces confident-sounding text that may be wrong. Human oversight is not optional here.
2. Manage client relationships. The trust that drives legal instructions is built through human interaction. AI can support client service — it cannot replace it.
3. Guarantee accuracy without verification. AI tools hallucinate. They produce plausible text that is factually incorrect. Any AI-assisted output in legal practice needs a verification step before it goes to a client or a court.
4. Navigate professional conduct rules autonomously. Confidentiality, conflicts, privilege, regulatory obligations — these require human oversight. AI does not understand the ethical obligations attached to legal practice.

Where to start

The most common mistake is starting with the wrong question. "What AI tools should we be using?" invites vendor pitches and shiny-object syndrome. The better question is: "What are we spending the most time on that AI might handle better?"

Start with a single high-volume, repeatable task. Map how it currently works. Identify where the time goes. Then ask whether a tool exists that addresses exactly that bottleneck — not the whole process, just the bottleneck.

A structured AI Discovery Day is often the right first step: a focused session with a specialist who understands both AI capability and legal practice, aimed at producing a ranked list of use cases with a realistic view of what each would take to implement.