In May 2026, a freelance HR consultant walked into Wandsworth County Court in London to recover £7,000 in unpaid fees. Most of the legal work that got her there had not been done by a traditional solicitor. It had been handled by Garfield AI, an AI-powered law firm authorised by the Solicitors Regulation Authority. Garfield prepared the pre-action correspondence, issued proceedings, dealt with a counterclaim, prepared four witness statements and assembled the trial bundle. A human barrister handled the courtroom advocacy. The consultant won.
Three months later in Sydney, a very different AI-assisted legal story played out. A former ALDI Stores Australia employee pursued an unfair-dismissal claim at the Fair Work Commission even though he had been dismissed three days short of the six-month minimum employment period. His AI-generated submissions repeatedly focused on the wrong legal issue, despite the Commission warning him in plain English that his case was effectively doomed. He eventually discontinued and was ordered to pay $1,230 towards Aldi’s legal costs.
Two cases. Same technology family. Opposite outcomes.
That contrast captures something important about the future of law. AI is making legal information and legal production dramatically more accessible. It can help people draft, research, interpret, organise and litigate. But access to an answer is not the same as knowing whether that answer is right, relevant, prudent or worth acting on.
Which is why I suspect AI will not simply make lawyers cheaper or fewer. It will change what clients are actually paying them for. As legal knowledge becomes easier to access, the premium will increasingly move to judgement.
Legal knowledge is losing its scarcity
For most of modern legal practice, several different forms of value have been bundled together. Lawyers had access to the law, knew how to interpret it, could draft the documents, understood the process and were able to advise a client on what to do next. Clients generally had to buy the bundle. AI is starting to pull those layers apart.
Consumers can already upload a contract and ask for a plain-English explanation. They can generate a first draft of a letter of demand, build a chronology, compare clauses or ask a chatbot what rights might apply to a dispute. Tools such as Legalese Decoder explicitly market themselves on making legal documents understandable to non-lawyers. At the other end of the market, law firms are deploying specialist systems such as Harvey, CoCounsel and Copilot to accelerate research, due diligence, drafting and document review.
The pace of adoption is striking. The 2025 ALPMA: Australasian Legal Practice Management Association and Dye & Durham APAC Changing Legal Landscape report surveyed 181 firms across Australia and New Zealand and found more than 90 per cent of law-firm leaders were experimenting with generative AI, while 70 per cent reported regular use. Use of tools such as Copilot, CoCounsel and other generative-AI systems had risen from 18 per cent of surveyed firms in 2024 to 64 per cent in 2025.
The important point is not that AI is becoming another piece of software in a lawyer’s toolkit. It is that some of the information advantage that once sat almost exclusively with the profession is moving into the hands of clients as well.
That changes the starting point of the relationship. Increasingly, a client may not arrive asking, ‘What does the law say?’ They may arrive saying, ‘ChatGPT says I have a strong case. Is it right?’ That sounds like a subtle distinction. It is not.
An answer is not the same as judgement
The legal system is already producing examples of people receiving plausible-sounding output and mistaking it for competent advice. In 2025, the Fair Work Commission rejected an application lodged 919 days late after the applicant relied on a suggestion from ChatGPT. Deputy President Tony Slevin described the proceedings as ‘hopeless’ and warned of the danger of relying on AI for legal advice. In another Australian matter, the NSW District Court encountered a self-represented litigant relying on a judgment that was entirely imaginary.
The problem is not always hallucination. Sometimes every case cited is real and the reasoning is still useless because it answers the wrong question. That was essentially the problem in the Aldi matter. The AI-generated submissions contained legal material that was more or less correct in places, but irrelevant to the threshold issue that decided whether the claim could be brought at all.
This is where the value of experienced human judgement becomes clearer. Knowing the rule is one thing. Knowing which rule matters, what facts are decisive, when a technically available argument is strategically foolish, and when the smartest legal move is not to fight at all are different capabilities.
A particularly revealing example surfaced in a Delaware Court of Chancery decision involving South Korean games publisher KRAFTON Inc. and the former owners of Unknown Worlds, the studio behind Subnautica. Krafton had agreed to an earnout worth up to US$250 million. When internal projections suggested that payout was likely to be triggered, the court found that Krafton’s CEO turned to ChatGPT after being warned by his own legal team that dismissing the studio leaders would probably not eliminate the earnout and could expose the company to litigation and reputational risk.
ChatGPT initially told him the earnout would be difficult to cancel. Yet at the chatbot’s suggestion, Krafton created an internal ‘Project X’ taskforce and later followed much of a ChatGPT-generated no-deal strategy. The court ultimately found Krafton had breached the transaction agreement, reinstated the studio CEO’s operational control and extended the earnout period.
The lesson is not that ChatGPT masterminded the failure. The judgment itself makes clear that the system initially warned of the difficulty. The more interesting lesson is that AI can rapidly supply plans, rationales and options to someone who is already motivated to pursue a particular course. In that setting, the most valuable adviser may be the one willing to challenge the premise rather than optimise the plan.
A phrase from a recent conversation with lawyers has stuck with me: AI can increasingly tell a client what they could do. A good lawyer earns their value by helping a client decide what they should do.
Law firms are already moving beyond the chatbot
It would be easy to read the cautionary cases and conclude that the profession should keep AI at arm’s length. The opposite is happening.
Maddocks provides a useful Australian example. After an 18-month AI program and a structured Harvey pilot across 13 practice teams, the firm announced in June 2026 that it was rolling Harvey out enterprise-wide. During the pilot, 70 per cent of participating lawyers were running queries daily and 88 per cent returned week after week. The intended use was not novelty prompting. Maddocks identified document review, due diligence and drafting as workflows to accelerate, while retaining professional oversight.
Herbert Smith Freehills Kramer has gone further in workplace investigations. Its ‘Alex the Investigator’ system was designed around a hybrid model rather than a pure AI replacement. The firm says the tool can compress investigations that previously took four to six weeks into roughly one week, while retaining humans in the loop to protect the employee experience and exercise judgement. That timing is a firm-reported result rather than independent benchmarking, but the model itself is instructive.
The next phase of legal AI will not be defined by lawyers occasionally opening a chatbot. It will be defined by workflows that are rebuilt around machines doing more of the searching, sorting, summarising, comparing and first-pass drafting, with humans positioned where judgement, accountability, advocacy and trust matter most.
That distinction matters because it changes what ‘productivity’ means. If AI reduces a ten-hour task to two hours, the opportunity is not simply to do the same work faster. It is to rethink who does what, what the client pays for, how quality is checked and where human expertise should be concentrated.
The billable hour has an AI problem
For law firms, this creates an awkward economic question. The billable hour assumes that time is a reasonable proxy for value. AI weakens that relationship. If an AI-assisted lawyer can complete in 30 minutes what previously took five hours, efficiency becomes financially self-defeating under a pure hourly model. The firm has created more value for the client and, on paper, less revenue for itself.
Columbia Business School professor Rita Gunther McGrath has argued that this is one reason AI makes time-based professional-services pricing increasingly untenable. The Wall Street Journal has also documented the messy transition already under way across consulting, where firms are experimenting with fixed fees, subscriptions and outcome-based pricing while struggling with the risks of abandoning a model around which staffing, utilisation and partner economics have been built for decades.
The legal market is showing similar movement, although predictions of the billable hour’s imminent death are premature. In the ALPMA survey, fixed-fee and alternative-pricing arrangements rose from 45 per cent of firms in 2022 to 64 per cent in 2025. Yet time recording remains deeply embedded in legal practice, and large firms still rely heavily on hourly billing.
What AI does is expose the weakness in the equation. Clients do not really want hours. They want risk reduced, problems resolved, deals completed, disputes avoided and decisions improved. Hours were simply an observable unit that both sides learned to use.
As the time required to produce legal work falls, law firms will be forced to articulate value more clearly. That may mean fixed fees for defined work, retainers for ongoing access, subscriptions for continuous monitoring, or pricing linked to outcomes where regulation and professional obligations allow it.
It may also explain another striking finding in the ALPMA data. Access to senior lawyers rose from 50 per cent of firms identifying it as a client value-add in 2019 to 72 per cent in 2025. At precisely the moment technology is making legal production easier, clients appear to be placing more value on direct access to experienced human judgement.
AI may change the leverage model before it changes lawyer numbers
This also complicates the most common question about AI and law: how many lawyers will it replace?
There are good reasons to expect some tasks to require fewer human hours. Research, document review, first drafts, comparison and summarisation are obvious candidates. But jobs are not tasks, and cheaper legal production can create effects in the other direction.
If the cost of pursuing a £7,000 debt falls from several thousand pounds of legal fees to a few hundred, work that was previously uneconomic can enter the legal system. If corporate teams can review more contracts or investigate more issues at the same cost, the volume of matters can rise. And if AI output requires checking in high-stakes contexts, some of the time saved in production reappears as verification, supervision and risk management.
So far, the evidence does not support a simple story of collapsing legal employment. ALPMA found 73 per cent of surveyed firms planned to increase solicitor headcount in FY26. Separate Australian Financial Review Law Partnership Survey reporting found graduate numbers and non-partner lawyer numbers rising even as AI became standard for more junior-level work. That does not prove the pattern will continue, but it is an important reminder that technological efficiency and employment do not always move in opposite directions.
The deeper issue may be what happens to the traditional law-firm pyramid. For decades, junior lawyers learned by doing large volumes of research, review and drafting, while firms earned margin on that work. If machines absorb more of those tasks, firms face a development problem as much as a staffing problem: how do future senior lawyers acquire the pattern recognition and judgement that comes from years of exposure to real matters?
That question is still unresolved. But it suggests that the winning firms will not be those that simply remove juniors from the equation. They will be the ones that redesign how juniors learn, how work is supervised and how human expertise compounds when the repetitive parts of the job no longer take as long.
The lawyer’s premium is moving upstream
For clients, the practical change is already visible. AI makes it easier to arrive at a legal conversation informed. It can help organise facts, decode documents and generate useful first drafts. In many circumstances, lawyers would be sensible to encourage that rather than resist it.
But the closer a matter gets to consequential action, the more the unresolved human questions begin to matter. Is this technically correct advice actually relevant to my circumstances? What is the commercial cost of being right? How will the other side react? What precedent will this set? Is litigation worth the distraction? What would a regulator, employee, customer or board think if this became public?
Those are not merely information problems. They require context, experience, timing, trade-offs and ownership.
The same principle applies as agentic AI becomes more capable of taking actions rather than simply producing text. Australian legal academics have pointed out that an AI agent is not itself a legal person. If a business deploys one and it causes foreseeable harm, responsibility does not disappear into the machine. The user, deployer, developer or other actors may still have legal obligations depending on the circumstances. More autonomy for software may therefore create more demand for human clarity about accountability.
This is why I think the most important shift in law is not from human lawyer to AI lawyer. It is from legal knowledge to legal judgement. Information is becoming abundant. Drafting is becoming faster. First-pass analysis is becoming cheaper. None of that makes expertise irrelevant. It changes the location of its value.
The lawyer who built a career on being the person who could find the answer may feel exposed. The lawyer who can test the answer, understand the situation around it, tell a client when not to act and take responsibility for the recommendation may become more valuable.
In other words, AI is not removing the need for legal expertise. It is forcing the profession to be much clearer about which parts of that expertise were scarce because information was hard to access, and which parts remain scarce because judgement is hard to automate.
That may be uncomfortable for a profession whose economics have long rewarded time and production. But it is also an opportunity. When almost anyone can generate a legal answer, the premium shifts to the person who knows what to do with it.
Michael McQueen is a globally recognised trend forecaster, change strategist and keynote speaker.
A bestselling author of 10 books, Michael’s latest release was named by Malcolm Gladwell and Adam Grant as one of the top five must-read new leadership books. He is a sought-after media commentator, with his insights regularly featured in Forbes, The Guardian, and CNN.
To find out more about Michael and his work, click here.
NOTES
Australasian Legal Practice Management Association and Dye & Durham. Trailblazing Tomorrow: People, Technology and Law on the New Frontier: 2025 Changing Legal Landscape. 2025. https://www.alpma.com.au/wp-content/uploads/2025/09/250915_DD_Alpma-2025-Report.pdf
Australasian Legal Practice Management Association. “Lawyers Face Work-Life Paradox as AI Adoption Soars Across the Profession.” September 16, 2025. https://www.alpma.com.au/news/lawyers-face-work-life-paradox-as-ai-adoption-soars-across-the-profession/
Solicitors Regulation Authority. “SRA Approves First AI-Driven Law Firm.” May 6, 2025. https://www.sra.org.uk/garfield-ai
Garfield AI. “AI Lawyer in Historic First Court Trial Win.” June 22, 2026. https://www.garfield.law/press/garfield-ai-wins-first-court-trial-with-regulated-ai-lawyer
Moloney, Charlie. “HR Consultant Wins English Court Case Using AI Lawyer in Apparent Legal First.” The Guardian, June 22, 2026. https://www.theguardian.com/technology/2026/jun/22/artificial-intelligence-law-firm-wins-court-case-in-england-for-first-time
Fair Work Commission. Deysel v Electra Lift Co. [2025] FWC 2289, August 8, 2025. https://www.fwc.gov.au/documents/decisionssigned/pdf/2025fwc2289.pdf
Muroi, Millie. “Sacked Worker Forced to Pay for Using ‘Plain Wrong’ AI in Landmark Case.” The Sydney Morning Herald, August 19, 2026. https://www.smh.com.au/business/workplace/sacked-worker-forced-to-pay-for-using-plain-wrong-ai-in-landmark-case-20260819-p60puw.html
Whitbourn, Michaela, and Stephen Brook. “Sydney’s Judge Judy Takes Aim at ‘Imaginary’ AI Judgment.” The Sydney Morning Herald, September 3, 2025. https://www.smh.com.au/cbd/sydney-s-judge-judy-takes-aim-at-imaginary-ai-judgment-20250902-p5mrpw.html
Fortis Advisors, LLC v. Krafton, Inc., C.A. No. 2025-0805-LWW (Del. Ch. March 16, 2026). https://courts.delaware.gov/Opinions/Download.aspx?id=392880
Maddocks. “Maddocks Advances Its AI Strategy by Rolling Out Harvey Enterprise-Wide.” June 23, 2026. https://www.maddocks.com.au/insights/maddocks-advances-its-ai-strategy-by-rolling-out-harvey-enterprise-wide
Gonski, Michael. “I Am Very Excited to Introduce to You Today ‘Alex the Investigator.’” LinkedIn, October 2025. Firm-reported description of Herbert Smith Freehills Kramer’s AI-enabled investigations tool. https://www.linkedin.com/posts/activity-7384707366635802628-lWFH/
McGrath, Rita Gunther. “Say Goodbye to the Billable Hour, Thanks to AI.” The Wall Street Journal, December 4, 2025. https://www.wsj.com/tech/ai/ai-goodbye-to-billable-hours-cba198fe
Maurer, Mark. “Inside Consultants’ Messy Shift From Hourly Billing.” The Wall Street Journal, June 26, 2026. https://www.wsj.com/cfo-journal/inside-consultants-messy-shift-from-hourly-billing-7bd9b802
Tadros, Edmund, and Hannah Tattersall. “Law Firms Take More Graduates Even as AI Does More Grunt Work.” Australian Financial Review, June 30, 2025. https://www.afr.com/companies/professional-services/law-firms-take-more-graduates-even-as-ai-does-more-grunt-work-20250626-p5malv
Shepherd, Tory. “AI Agents Aren’t Legally Responsible for Any Harm That They Cause, Experts Say. So Who Is?” The Guardian, August 13, 2026. https://www.theguardian.com/technology/2026/aug/13/ai-agents-arent-legally-responsible-for-any-harm-that-they-cause-experts-say-so-who-is
