AI and the Changing Shape of Legal Practice – 2026

If your firm’s AI conversation is still mainly about whether lawyers should use ChatGPT/Claude/[insert your preferred frontier model here], how to write better prompts, or privilege concerns, don’t worry – you’re in good company.

A small problem is that the conversation elsewhere appears to have already moved on.

That was one of my clearest impressions from the Australian LawTech Summit I recently attended on the Gold Coast.

At the conference there was a palpable energy – coupled with a decent dose of trepidation – about what firms are beginning to achieve with AI and what that means for supervision, pricing, training, operating models and the delivery of legal services. One session also raised a more unsettling question: is the person instructing you on a video call really your client?

Key takeaway? Some of what is already happening is further along than many practitioners may realise.

A few things stood out.

Not All AI Use Is Equal
Figures cited at the Summit suggested that more than 80% of legal professionals are now using AI.

But “AI adoption” can mean almost anything.

A lawyer who occasionally asks AI to polish an email and one who uses it throughout the day for research, drafting, document review and analysis may both be counted as AI users. On paper they look the same. In practice, they are operating in different worlds.

A useful distinction raised at the Summit was between:

• use – whether AI is being used, how often and for what;
• capability – whether the practitioner can frame tasks well, provide context, recognise limitations, test outputs and spot when something does not look right; and
• execution – whether the firm has moved beyond individual experimentation to approved tools, workflows, governance and training.

It is quite possible to score well on the first and poorly on the other two.

A related issue is that firms are still grappling with what AI means for their operating and commercial models – how to measure efficiency gains and track outcomes, and how to maintain revenue when the same work takes less time.

From Prompts to Workflows
Firms further along are increasingly talking about agents, workflows, standing instructions and context.

An AI agent can be given a defined role, detailed instructions, approved sources and tools, limits on what it may do, rules about what it must check and points at which it must escalate to a human.

Put several steps together and the change becomes more interesting.

A workflow might review source documents, extract facts, prepare a chronology, identify issues, produce a first draft, check that draft against the source material and flag omissions or unsupported propositions – all before the work reaches the lawyer.

But instructions are only part of the equation.

Think of a capable graduate on their first day. Give them a file with no background about the client, matter history, precedents or how the firm likes things done, and even good work may miss the mark.

AI has much the same problem.

Firms are therefore increasingly giving AI richer and more controlled context: precedents, drafting style, policies, preferred sources, client and matter information, risk settings and standing instructions. That context is becoming something firms deliberately build and maintain, much as they have traditionally built precedent and knowledge banks.

So the question is no longer only, “Which AI tool should we use?”

It is also, “What are we giving it to work with?”

Who Checks The Work?
We have all heard the warnings about hallucinations. They are justified. AI-generated work must be checked.

What is changing is how that checking can occur.

One agent might perform a task. Another can check the result against the source material. A third can look for omissions, inconsistencies or unsupported propositions. Further checks can test the work against legal, procedural or firm requirements.

The first several passes of verification may therefore occur before the lawyer even sees the work.

None of this shifts professional responsibility. The practitioner remains responsible for work relied upon, provided to a client or put before a court. Critical propositions and authorities still require appropriate verification against reliable primary sources.

But in the very near future, supervision may involve overseeing the reliability of a system of work, rather than checking every step undertaken within it.

Capability Without a Big Budget
There was also encouraging news for smaller firms. Serious AI capability does not necessarily require a serious technology budget.

Specialist legal AI platforms offer real advantages, including curated legal content, integrations, audit trails and additional contractual or governance protections. For some firms they may absolutely be the right answer.

But sophisticated capabilities are also appearing in mainstream platforms such as ChatGPT and Claude, including through desktop applications and more controlled, permission-based environments. These can support integrations, standing instructions, agents, document analysis, context-rich workflows and layered verification.

For smaller firms, the more important investment may sometimes be in the environment around the AI:

• cybersecurity and access controls;
• information governance and document management;
• data classification;
• good precedents and knowledge resources;
• reusable instructions and workflows; and
• staff capability and training.

Before any platform is used for client work, firms also need to understand what happens to the information being put into it, including arrangements concerning confidentiality, data use, retention, access and storage.

In fact, for some firms the most valuable AI project may begin without AI at all: clean up the documents, sort out permissions, classify the information and decide what any system should actually be allowed to see. And work out a good governance layer.

Not quite as exciting as an AI agent. Possibly more important.

Rethinking The Billable Hour
Suppose a task that previously took five hours can now be completed properly in one or two.

Excellent.

Now what?

Is the benefit passed through to the client? Retained by the firm? Reflected in higher volumes, better margins or faster service? A combination of all?

These questions go beyond pricing. They go to the business model itself.

If lawyers are rewarded primarily for time recorded, the person who finds a faster and better way to do the work can paradoxically look less productive. A firm may therefore be encouraging its lawyers to embrace AI while its performance measures may be quietly incentivising them not to.

The value to the client may lie in the outcome, speed, certainty and convenience, but also in the lawyer’s judgement, understanding of the client and trusted adviser relationship – rather than simply in the number of hours spent on the work.

That may support a broader mix of fixed, capped, value-based, subscription and time-based pricing depending on the work.

The more useful question may therefore be less “How many hours did this take?” and more “What value did this create?”

The billable hour is not disappearing tomorrow. AI is, however, making these contradictions increasingly difficult to ignore.

How Will Junior Lawyers Learn?
This was one of the harder questions raised at the Summit – and one that, it would appear, remains largely unanswered.

Junior lawyers have traditionally developed expertise by doing the work: research, document review, chronologies, due diligence, first drafts and source checking.

Unfortunately – or fortunately, depending on which hat you are wearing – those are precisely the kinds of tasks AI is increasingly capable of performing.

If AI produces the first chronology, research note or draft, a junior lawyer may encounter a polished output without doing the underlying work through which pattern recognition and judgement have traditionally developed.

And there is the paradox.

The better AI becomes at legal work, the more important strong underlying legal judgement becomes, because lawyers need that expertise to recognise when AI is wrong.

Summit speakers described this as “AI intuition” – the ability to recognise that an answer is not quite right, that something important has been missed or that an apparently confident output warrants greater caution.

Basic familiarity with AI can develop quickly. That intuition takes longer. It depends on legal experience as much as technological capability.

So how do we develop it if we automate away some of the work that traditionally taught it?

There was no neat answer. But firms may need to become much more deliberate about preserving opportunities for junior lawyers to undertake, understand and interrogate foundational legal work, rather than simply reviewing an AI-generated product.

“We Trained Machines To Think Like Lawyers. Now What Do We Train Lawyers to Do?”
That was the title of one of the Summit’s most thought-provoking sessions, presented by Dr Kellie Nuttall, Executive Director of Future Secure AI.

It neatly captured the bigger issue.

As AI becomes embedded in legal work, lawyers will need capabilities beyond simply knowing how to operate the technology. They will need to supervise it, challenge it, recognise when something is wrong and know when human judgement must take over.

But there is another dimension to the training question. AI is not only changing how lawyers work; it is changing the environment in which they practise. That may require additional core skills to be developed through legal education, practical legal training and ongoing professional development – including assessing authenticity, understanding provenance and metadata, testing material against independent sources and recognising when specialist forensic expertise is required.

Which brings me to another question raised at the Summit.

What If The ‘Client’ Isn’t Your Client?
One Summit session brought that question into sharp focus, examining fabricated evidence, deepfakes and the increasingly difficult problem of whether what we see and hear can be trusted.

The risk is not theoretical.

In one widely reported case, an employee of a large international company joined a workplace video conference with people he believed were its CFO and other senior colleagues. Following the meeting and subsequent instructions, he made 15 transfers totalling about HK$200 million.

Turns out, the people on the call were extremely convincing AI-generated deepfakes.

And by the way, did you know that audio and video can now be manipulated in real time – including during live calls and even CCTV feeds? Edit in a comment, edit out a person … you get the gist.

Issues are also appearing in legal proceedings. In a 2025 Californian case discussed at the Summit, video testimony was submitted from a witness who had never actually given it. The fabrication was, at the time, relatively crude and was detected by the judge. But the technology has moved quickly since then.

The point was that this fake was caught. The next one may not be.

There is also a reverse problem. As fabricated material becomes easier to create, genuine evidence may increasingly be challenged on the basis that it could have been generated or manipulated.

For lawyers, that brings questions of authenticity, provenance, metadata and independent verification into sharper focus. And it is not confined to litigation.

Sometimes the question may be much more basic: is the ‘client’ giving instructions on the video call actually the client?

A Changing Professional Landscape
AI is becoming much more than a better drafting tool. It is beginning to change the environment in which legal work is done – including how work is organised, supervised, measured and priced.

The fundamentals have not changed: competence, confidentiality, supervision, independent judgement and professional responsibility remain central.

What is changing is what lawyers may need to do to discharge those responsibilities well.

For the profession, the challenge is therefore no longer simply learning how to use AI. It is working out what capable, responsible legal practice looks like when AI becomes part of the way legal work is done.

And we probably need to work that out reasonably quickly.

A Final Parting Thought
At the Summit, one speaker explained exponential change in wonderfully simple terms.

Take 30 ordinary one-metre steps and you travel 30 metres.

Make those steps exponential and, by the 30th step, you have travelled the equivalent of around the planet 26 times.

Whether, in reality, technological development is following quite so neat a trajectory, as stated by many, is another question. But the illustration stuck with me.

At the present rate of change, what will we be talking about at the next Summit?

I’ll report back then.

September 2026

Francesca Beattie
Deputy Executive Director
The Law Society of Tasmania

What is happening in Tasmania?

The Law Society is working with the University of Tasmania, the Centre for Legal Studies and representatives of the profession through the Tasmanian Legal Community Generative AI Working Group.

The Group is examining what AI capability may need to look like across legal education, practical legal training and ongoing professional development – including the difficult question of how future lawyers develop the underlying legal skills and judgement required both to practise competently and to use these technologies well.

Its current survey of the Tasmanian profession (closing 5 October) will provide an initial evidence base on AI use, capability, barriers, governance and the extent to which AI is becoming integrated into legal practice.

Those results will help inform the next phase of the Working Group’s work, including engagement with the profession, educators and other stakeholders about what practical responses, training and support may be required.