The Last Human Lawyer: What Legal Tech Still Cannot Do.

Legal technology has become wonderfully good at making lawyers feel both powerful and faintly redundant before breakfast. It can review a disclosure bundle before the coffee machine has emotionally committed to the day. It can draft a service-level agreement without developing a grudge against the client. It can identify patterns, flag risk, summarise evidence, compare clauses, and produce a first draft that looks suspiciously like the work of an exhausted associate who’s had 3 coffees and a private argument with a semicolon.

After a series about algorithms, evidence, and acceleration, this is the necessary pause. Not a retreat from technology. Not a scented-candle protest against progress. Just a hand on the table, a small clearing of the throat, and the question nobody in a boardroom full of dashboards especially wants to ask: if AI can draft the NDA, write the article, organise the diary, summarise the evidence and identify legal risk faster than a human, what exactly are all those humans now for?

The beautiful lie of replacement

The loudest sales pitch in legal tech has always been efficiency. Faster research. Cleaner documents. Better matter visibility. Less human error. Fewer hours wasted on work that slowly drains the soul through the left eye. All true. All useful. All overdue. But somewhere between the automated NDA and the AI-assisted litigation strategy memo, we started smuggling in a more dangerous idea: that because machines can perform parts of legal work, they may soon replace the human centre of legal judgment.

That’s where the conversation becomes less about software and more about grief. Lawyers, paralegals, legal secretaries, writers, journalists, and support staff aren’t only asking whether their tasks will change. They’re asking whether their usefulness has an expiry date. They’re watching the ordinary work that once paid salaries and built careers – the NDA, the employment contract, the first draft, the research notes, the diary, the filing, the client update – disappear into tools that can produce it in seconds. And they’re asking, quietly and with impressive professional composure, whether the thing they spent years becoming still matters.

The answer cannot be a smug motivational poster about “embracing disruption” (eye roll). Nobody needs another panel discussion where 5 people in navy jackets explain that AI will “free lawyers to do higher-value work”. Do lawyers actually want a career made up only of the hardest, highest stakes matters, with no breathing room and no simpler work to balance the load? And what happens to the people whose bread and butter was precisely that simpler work? “Higher-value work” sounds reassuring in a strategy deck. It sounds rather different when the work that vanished was paying your bond.

But beneath the strategic language is a question far more intimate than productivity: where do people go when the work that gave them income, confidence, status, and a sense of usefulness disappears? Not every displaced employee becomes an AI strategist. Not every legal secretary becomes a data analyst. Not every junior lawyer leaps joyfully into complex advisory work. Some people simply discover that the profession still values the outcome but no longer needs them to produce it. That’s not disruption on a slide. It’s a person sitting in a parked car before going home, trying to work out how to tell their family that efficiency has acquired their salary.

There’s also a form of support firms rarely put in the business case because it cannot be expressed neatly in billable hours: helping people survive the emotional shock of becoming professionally unnecessary. Confidential counselling, peer support, financial-planning assistance, extended medical cover and time to search for work are not decorative benefits. They recognise that displacement can fracture identity as much as income. Leaders don’t need to promise that every job can be saved. But they do need to tell the truth early, listen without defensiveness and resist describing another person’s loss as an exciting efficiency gain.

The machine can find the pattern. It cannot read the room.

AI is particularly good at spotting patterns. In law, that’s not a novelty. It’s useful, serious work. Precedent, deadlines, inconsistent clauses, the one sentence in 800 pages that changes everything – these are exactly the things machines can help us find. Thomson Reuters reported in 2026 that many legal professionals expect AI to reshape the profession, while worrying about what happens to independent judgment along the way. There it is: the promise and the problem, sharing a desk.

A machine can show you where the risk is. It can’t always tell you what the risk means inside a particular organisation, on a particular Tuesday, with a particular CEO who has started using phrases like “commercially robust” in the same tone other people use when reversing over a traffic cone.

Law is rarely only about the words on the page. It’s about why those words were chosen, who’s avoiding eye contact when they’re discussed, which internal department has gone mysteriously silent, and why the client’s confidence has the brittle texture of a cheap biscuit. That’s context. And context is where the human lawyer still earns their keep.

The death of easy work has consequences.

There’s a fashionable comfort in saying automation will remove drudgery and leave humans with only complex, meaningful work. Lovely. Inspirational. Also slightly deranged if taken literally. Routine work wasn’t only profitable; it was formative. It was where junior lawyers learnt how contracts breathed. It was where they saw how clauses collided, how instructions mutated, and how a “quick amendment” could become a litigation grenade with a lovely font.

If machines take over too much of the bottom of the pyramid, the profession must answer a serious training question: where will judgment come from? You cannot outsource the apprenticeship of attention and then act surprised when future lawyers struggle with nuance. You cannot remove the slow work that teaches people how legal consequences accumulate and then demand mature strategic instincts at partner level, preferably by Thursday.

This isn’t an argument for preserving inefficiency as a heritage site. Some legal admin deserved to die peacefully in its sleep years ago. But if we automate the grind, we must intentionally rebuild the learning it once provided. The future lawyer cannot be trained by vibes, prompts, and emergency supervision. As chaotically fun as that sounds. They need structured exposure to consequence, accountability, and the discipline of verifying what appears plausible.

The real danger isn’t that every job disappears overnight. It’s that working life gets hollowed out from the middle. Routine contracts, first drafts, scheduling, checking, and follow-up once formed a ladder. People learnt the work, earned their place, and became trusted by climbing it. If automation removes the rungs, telling everyone to aim for the top isn’t a plan. Firms will have to build another way up – one that lets people grow with the technology instead of standing beside it while it gets better at the work that once paid them.

The people whose tasks are being automated should help choose the tools, test the outputs, identify failure points, and decide where human review remains non-negotiable. That’s not sentimental consultation. The legal secretary who has rescued 400 matters from diary chaos and the paralegal who can spot a missing exhibit in seconds carry expertise no workflow diagram captures. Use that knowledge as design input and both the employee and the system have a better future.

And please, not another lunchtime webinar called Thriving Through Change, followed by a link to a course nobody has time to finish. Firms need to show people where the next job is. Junior lawyers can learn to interrogate AI-generated work, interview clients, and exercise judgment earlier. Support staff can become matter coordinators, workflow specialists, knowledge stewards, and quality controllers. Writers and researchers can lean harder into investigation, interviews, interpretation, and editorial instinct. The work may change. The route into it cannot be imaginary.

The oldest legal skill is still courage.

The most important thing legal tech cannot automate is not drafting. It’s not research. It’s not even negotiation, although watching an AI try to navigate a hostile settlement discussion would at least provide premium entertainment.

The thing it cannot automate is moral courage.

A lawyer’s most valuable sentence is often not “we can do this”. It’s “we shouldn’t”. That sentence is expensive. It costs political capital. It can irritate the person paying the invoice. It can make a room colder than a Cape Town conveyancing department in July. But it’s the sentence that separates legal service from legal obedience.

Generative AI responds to prompts. Lawyers respond to duties. That distinction matters. The American Bar Association’s 2024 ethics guidance on generative AI emphasised competence, confidentiality, communication, and reasonable fees. South African legal commentary has made the same point in sharper local terms: AI can support legal practice, but practitioners remain responsible for the work, the verification, and the judgment.

In South Africa, the warnings are no longer theoretical. De Rebus has discussed both the promise and risk of AI-assisted legal work, including cases where courts cautioned against uncritical reliance on AI-generated authorities. The point is painfully simple: but plausible isn’t the same as true. Anyone who’s survived a commercial dispute, a committee meeting or a family WhatsApp group already knows this.

The machine doesn’t have a practising certificate, a reputation, a client relationship, a conscience, or the ability to wake up at 02:17 wondering whether the advice wasn’t only defensible but right. Human lawyers do. That burden isn’t a bug in the legal system. It’s the point.

Clients do not only bring facts. They bring fear.

Corporate clients often arrive with documents, timelines, and instructions. They also arrive with fear. Fear of exposure. Fear of losing control. Fear of being found out. Fear of having to admit that the clean strategy deck is sitting on top of a human mess that has been quietly fermenting for years.

A dashboard can organise that fear. It cannot absorb it. A chatbot can respond politely. It cannot sit with a client while the room goes silent after someone finally tells the truth. A risk engine can rate exposure. It cannot decide whether a frightened executive needs reassurance, confrontation, containment, or the kind of blunt advice that feels cruel for 5 minutes and merciful for the next 5 years.

The Law Society has noted that technology can strengthen client relationships when it reduces administrative burden and allows lawyers to focus on what matters: the client. That’s not a soft point. It’s commercial reality. Trust isn’t ornamental. It’s a risk-management tool. Clients tell the truth to people they trust, and legal advice improves dramatically when it’s based on the truth rather than the curated executive version of events.

Every lawyer knows the moment. The client has technically answered the question, but not really. The words are there. The answer is not. Something flickers. A pause lands too heavily. A detail is over-explained. Suddenly the real matter enters the room, carrying a laptop bag and wearing the expression of a person who hasn’t slept properly since the board pack went out.

That’s where lawyering lives. Not in mystical human exceptionalism. Not in pretending lawyers are morally superior beings who’ve never hidden behind “per my previous email”. Lawyering lives in the ability to read facts and people at the same time, then give advice that recognises both the legal position and the fragile human being who must act on it.

The human lawyer is not anti-technology.

None of this means we should clutch our annotated statutes to our chests and hiss at innovation from the corner office like an agitated cat. The anti-tech lawyer is not noble. Often, they’re just slow, expensive, and strangely proud of making everyone email the same PDF 9 times.

The better future is not humans on one side and technology on the other, glaring at each other across a boardroom table – “I see you buddy”. It’s people using good tools without being quietly written out of the story. Let machines find patterns, organise matter data and clear away the administrative sludge that has ruined too many evenings. But don’t keep a small circle of senior decision-makers and call the result “human-led”. Lawyers, writers, paralegals, and support staff must still have room to learn, contribute, and share in the value the technology creates.

Firms cannot sign the licence agreement and leave the human consequences in HR’s inbox. Look at the roles before automating them. Tell people what’s changing before the office rumour mill does. Pay for retraining during working hours. Move people internally where there is real work to move them into. Create proper entry points into AI review, quality control, legal operations, and client service. And where no meaningful role remains, offer time, career support, and fair severance. If technology saves the firm money, the person whose job disappeared shouldn’t be left to finance their own rescue. That’s just not fair.

Support must begin before a role disappears, not after the farewell email is drafted. No surprise restructures. No pretending a disappearing role is an exciting development opportunity. And no reducing a person to a task inventory. Institutional memory, client knowledge, judgment, relationships and the thousand invisible acts that keep a firm functioning are part of the value too.

Human beings must still own the judgment, the ethics, the strategy, and the final answer. Legal teams are under pressure to prove that AI creates value. Fair enough. But accuracy, security and supervision don’t disappear because the software is impressive. Use the machine. Just don’t point at it when the advice goes wrong.

Because the ugliest risk in legal tech isn’t that AI will become too powerful. It’s that humans will become too passive. That we’ll accept machine output because it looks polished. That we’ll confuse fluency with truth. That we’ll stop asking the annoying, necessary questions because the answer arrived instantly and in a tone that sounded terribly confident.

What remains

So, after the contracts have been drafted, the evidence sorted and the risk neatly colour-coded, what’s left for us?

• Judgment, for one: the messy weighing of law, fact, timing, politics, consequence, and human frailty, followed by a decision somebody must be willing to own.
• Ethics too. Not the laminated policy in reception, but the hot, unpleasant moment when a professional has to say no.
• Then there’s context: the odd, local, political, personal texture that never fits neatly into a database and often decides the matter anyway.
• Negotiation still belongs here – not merely the exchange of positions, but the reading of ego, fear, leverage, fatigue, and that tiny change in a voice when someone realises their bluff has been called.
• So does trust. It’s built slowly, damaged quickly and impossible to automate without turning it into customer-service theatre, complete with hand puppets.
• And, finally, courage: disappointing a client in order to protect them; resisting speed when accuracy matters; telling a room full of powerful people that the legally clever option may still be morally stupid.

All blissfully human.

The last human lawyer

The last human lawyer isn’t the person who refuses AI. They’re not the grand traditionalist composing opinions by candlelight while muttering about young people and their portals. The last human lawyer is the one who uses technology without surrendering responsibility to it.

They understand that speed is useful but not sacred. They know a first draft is not a final judgment. They know that a confident answer can still be wrong, that a beautiful summary can still miss the wound, and that the client’s real problem may be hiding in the one fact nobody wanted to include in the instruction email.

Most of all, they remember that law isn’t only a system of rules and work isn’t only a collection of tasks. Both are places where people build identity, confidence, belonging and a sense that they matter. People turn to the law when something important is at risk: liberty, money, dignity, reputation, livelihood, legacy, sometimes sanity. The work may be commercial, technical, and heavily automated, but the stakes are still human – and so are the people doing it.

Legal tech will continue to improve. It should. We need it. We need better systems, safer data, faster workflows, and tools that stop talented professionals from being buried alive under admin. But if the future of law is only faster, it will be poorer. If it’s only automated, it will be colder. If it forgets the human lawyer, it will become very efficient at missing the point.

The machines may find the smoking gun. They may find it faster, cheaper and with fewer spelling errors than we’d like to admit. But they cannot yet carry the silence that follows. They cannot look across the table at a client who is frightened, cornered or lying, and decide what truth must be spoken next.

That is still our work. Not because we are faster. We are often not. Not because humans are wonderfully wise and machines are cold metal villains. We have all met humans. We remain relevant because consequence has to be understood, not merely calculated; because trust is not a transaction; and because courage does not arrive when somebody types the right prompt.

A person who fears becoming obsolete doesn’t need another speech about competing with the machine. They need somewhere real to stand. If we build the future of legal work around that, technology may not make people less valuable. It may simply force us to notice what their value was all along.

Contributed by AJS.

(Sources used and to whom we owe thanks – Thomson Reuters; The American Bar Association, for Formal Opinion 512; De Rebus and the authors of the South African generative-AI ethics guidelines; The Law Society of England and Wales, LexisNexis South Africa, Springer and Meritas).