The announcement gap
I read a press release earlier this year from a firm launching an AI programme. It sounded impressive. I know two people there. Neither had heard of it.
This is not cynicism. It is the actual state of things, and it is worth being clear-eyed about because the announcements are what everyone else is benchmarking themselves against.
Buying a licence is procurement. Changing how work gets done is management. Firms are considerably better at the first.
What is really in daily use
- Document review and first-pass diligence. The most mature use by some margin, and the one with the clearest return.
- Drafting. First drafts of routine documents, then a human doing what a human is for.
- Summarising. A long bundle, a long thread, a long call. Deeply boring and enormously useful.
- Transcription and note-taking. Underrated. It removes a real daily tax on fee earners and nobody has to be persuaded to like it.
Notice what is not on that list. Nothing here is a lawyer being replaced. All four are the parts of the day nobody went to law school for.
The work AI has taken is the work fee earners were doing at nine in the evening, because there was no other slot for it.
What is still pilot theatre
- A firmwide AI strategy with no named owner.
- Bespoke builds where something off the shelf would have done the job last quarter.
- Anything where the business case was written after the purchase.
- An innovation committee meeting monthly to review a pilot nobody is running.
None of this is a criticism of the people involved. It is usually a symptom of buying before deciding which problem you were solving.
Who is actually ahead
Not who you would expect. The firms moving fastest, in my experience, are small. Two to fifteen people.
Fewer systems to unpick, no procurement cycle, and the person deciding is the person doing the work.
Large firms have better tools and worse throughput. Small firms have worse tools and better habits. The second group is winning, for now.
The thing firms get wrong first
It is almost never the technology. It is client material going into a tool nobody approved.
This happens in every firm, at every size, and it happens because a fee earner under pressure found something useful. They are not being reckless. They are being efficient in the only way available to them, because the firm took nine months to decide on a policy.
Two things fix most of it. Give people a sanctioned tool good enough to be worth using, and say plainly what must never be pasted into anything else. A ban with no alternative does not stop the behaviour, it just stops you knowing about it.
The firms with a real problem coming are not the ones moving fast. They are the ones where nothing is approved, nothing is banned, and nobody has asked.
What it actually saves
Firms want a number here and the honest answer is a range, because it depends entirely on how much of the week was going on the four tasks above.
Where I have seen it measured properly, first-draft time on routine documents drops by more than half, and summarising work by considerably more than that. Diligence is the big one, and also the one where the saving is easiest to overstate, because the checking does not disappear - it moves.
What nobody measures, and should, is the time recovered at the wrong end of the day. A fee earner getting an evening back is not a line on a matter ledger, but it is the thing keeping people in the job.
What clients should be asking
If you instruct firms rather than run one, this is the part affecting you, and there are three questions worth putting straight to whoever pitches you.
- What do you use it for, specifically? A named task is a good answer. "We have an AI strategy" is not.
- What does a person always check before it reaches me?
- Where does my material go, and does it train anything?
A firm using these tools well will answer all three in about a minute. A firm using them badly, or pretending to use them at all, will get vague at the second question. That tells you a great deal, and it costs you nothing to ask.
What to do on Monday
- Pick one task, not one tool. Something done weekly, low risk, and irritating.
- Time it before. Time it after. Otherwise you are guessing, and so is everyone you report to.
- Decide what a human always checks, write it down, and hold to it.
- Tell people what you are doing. Quiet AI use inside a firm is how you end up with a problem nobody could see coming.
Frequently asked questions
How are law firms actually using AI?
Four things carry most of it: document review and first-pass diligence, drafting first passes, summarising long material, and transcription. Adoption is narrower than the announcements suggest and concentrated in tasks nobody enjoyed doing.
Is AI replacing lawyers in law firms?
No, and it is not the useful question. It is taking parts of the day - first drafts, summaries, note-taking. The judgement, the accountability and the client relationship have not moved.
Are small firms or large firms further ahead with AI?
On tooling, large firms. On actually changing how the work gets done, small firms are frequently ahead: fewer systems, no committee, and the person deciding is the person doing the work.
What does the SRA say about AI in law firms?
It has not prohibited it. The guidance keeps returning to the same point: you remain responsible for the work you produce and for client confidentiality, whatever produced the first draft.
Want to know what is worth adopting in your firm?
Silva's AI arm works with law firms and legal teams on exactly this - what to adopt, what to ignore, and how to put it into practice without a committee.