No Profession Is Safe, and the Lawyer Stands First in Line

How I can tell from the daily post of lawyers and authorities that two machines have long been talking to each other, why a well built prompt retrieves more legal knowledge than the colleague who reads into the file ten minutes before the hearing, and why I have argued my own cases myself for years, with the lawyer sitting beside me as the one who signs.

This morning I choked on my coffee, and not a little, but across the full width of the desk. The espresso took the shortest possible route from my mouth onto the open screen of my MacBook, spread itself evenly across the Retina display, found its way down between the keys and worked its path under the space bar, where no Apple technician on this earth will ever fully dry it out again. I watched that brown trickle seep away with the calm fascination of a man who has just understood that the real problem is not on the keyboard at all, but on the screen standing above it. The reason for the coughing fit was not the coffee, it was a single written pleading.

I have been reading legal correspondence and letters from the authorities for over two decades, professionally, daily, in a quantity that a normal person would regard as a carefully designed form of punishment. By now I recognise the rhythm of a pleading as reliably as a pathologist recognises a tissue under the microscope, by its grain, by its staining, by what is missing from it. That is precisely why I froze for a moment this morning, because in two letters that formally came from two different law firms, I read the same unmistakable heartbeat. It was not the style of two lawyers locked in a dispute with one another. It was the style of two machines politely quoting each other back and forth, and the human sitting between them does not even notice.

On one side the slightly fussy, endlessly balanced, list obsessed prose that a certain American language model leaves behind when you let it off the leash without clear instruction. On the other side the somewhat smoother answer, tuned more heavily toward coherence, which I would attribute to a different and likewise American model. Two systems feeding each other the ball, inside a dispute in which a human being pays the bill at the very end, and pays it for both sides at once. That is the most expensive conversation I have ever observed, and neither of the two interlocutors has a pulse. The coffee on the keyboard was the only living thing in that room with any measurable will of its own.

A Drive Through Italy and a Thought That Will Not Leave

I was on the road again yesterday, somewhere in Italy, on one of those routes you have driven so often that your hands hold the wheel while your head clears for the larger questions. People who know me personally know that my back rewards such drives in its own particular way, namely with strong pain that lingers for two or three days afterwards and refuses to be talked out of it by posture or by good intentions. This is not a complaint, it is a plain mechanical fact, and it has consequences that I rarely talk about. One of those consequences is that I can no longer work as a court appointed expert for the German justice system, and the reason is more banal than most people would assume. Every expert report is followed by the hearing, every hearing is followed by the journey, and every journey is followed by the back, which then sends me its invoice for the next few days. At some point you run the calculation cleanly once, and you find that the price no longer adds up.

On drives like that I think about where this world is heading, not out of melancholy, but out of professional curiosity, because I have spent my whole life taking systems apart until I understood how they actually work. The conclusion I keep arriving at on these drives is uncomfortable enough that I kept it to myself for a long time. We are living through the greatest transformation in the history of Homo sapiens, and we are living through it at a speed that this same Homo sapiens does not begin to grasp. The reason for that is anatomical, not a matter of character, and that distinction matters more than it first appears.

The brain with which a person answers an email at eleven at night is the very same organ that was optimised some 50,000 years ago to survive in a manageable group, to assess a danger in fractions of a second and then to settle back into calm. This brain is magnificent at spotting a predator at the edge of the forest, it is superb at reading a face in a crowd, and it is catastrophically bad at feeling an exponential curve that doubles itself within weeks. Exponential growth simply did not occur on the savannah, so evolution never built an organ for it. We therefore do not see the change, we merely hear about it, and that is an enormous difference, because hearing leaves us cold while seeing would make us run.

I am also fairly certain that governments do not have it on their radar either, or, which is more likely, that existing laws make it impossible for them to intervene logically and in time. Exactly once, a state recently did something that looks like a deliberate intervention, and it happened in the place where one would least expect it. A Chinese court in Hangzhou ruled at the end of April that a company may not dismiss an employee solely in order to replace him with an artificial intelligence, because the introduction of AI is an entrepreneurial decision and not an objective fundamental change of circumstances that would justify a termination (Fortune, 2026; NPR, 2026). It was not a prohibition statute, it was a single ruling, and that is exactly what makes it so remarkable. It shows that the most effective brakes in this whole process currently come not from the parliaments but from the courtrooms, and that ought to occupy us far more than it does.

Two Machines, One Client Who Pays for Both

The observation from the coffee table is not an isolated case, it is a pattern, and patterns are my actual trade. More and more often I read pleadings that a human has signed but has not thought. You recognise them by a series of symptoms, which I set out here with the clinical precision that the matter deserves, because the diagnosis gives me more pleasure than the finding. There is the conspicuous fondness for the rule of three, the eternal firstly, secondly and thirdly, where an angry human being would simply have said that the other side is in the wrong and left it there. There is the strange politeness that wraps every conflict in cotton wool, as though the pleading itself were afraid of annoying the judge. And there is the addiction to inflated connective phrases, this what is more and in addition and to round things off, which no living lawyer would ever voluntarily write three times in a single paragraph, because his own hand would fall asleep from the shame of it.

When I lay two such letters side by side, I no longer see a legal dispute, I see a game of table tennis between two servers. One model asserts something in impeccable office German, the other model contradicts it in even more impeccable office German, and in the middle sits a client who pays a fee for every one of these rallies, billed by the hour, as though a human being were still seated there wrestling with the problem. It is as if you let two pocket calculators compute against each other and then wired a consulting fee to both of them. The funny part is that it works, because nobody at the table wants to admit that he recognised the machine long ago. The less funny part is that we have arrived at exactly the point where it is no longer humans conversing but artificial intelligences, and that this happens in the very sphere that claims to be the most human of all, namely the search for justice.

What truly troubles me about this mechanical politeness is not the waste, absurd as that waste may be. It is the slow hollowing out of what a legal dispute is actually supposed to achieve. A proceeding is, at its best, a struggle between two human beings over a truth that neither of them possesses alone, and out of that struggle something is meant to emerge that neither of them held beforehand. When a machine sits at both ends, drawing on the same training data, then nobody is struggling any longer, and the two pleadings start to resemble each other like two drops from the same tap. The friction disappears, and with the friction disappears the insight that can only ever arise from friction. What remains is a perfectly worded draw, for which both sides have paid dearly, and the judge searches between two flawless texts for the one human error that will tell him which of the two still understood what the case was even about.

I am not reproaching anyone here for using these tools, that would be thoroughly dishonest of me. I use them myself, every day, and I use them well. I program faster with them, I find faults in unfamiliar systems in hours that used to cost me days, and I restore compromised servers while others are still opening the manual. The point I am making is an entirely different one. The point is that the ability which protected an entire profession for decades is right now turning into a commodity that anyone can pull from the wall socket for pennies. And when that happens, the question is no longer whether a profession changes, but how many people it can still feed at the end.

The Army of the Displaced, in Sober Numbers

So that nobody believes I am painting here with the broad brush of the doom merchants, I lay the numbers out openly, and I name them coldly, because they are cold. The International Monetary Fund established back at the beginning of 2024 that almost 40 percent of global employment is exposed to artificial intelligence, and in the advanced economies the figure rises to around 60 percent (IMF, 2024). The genuinely disturbing part of this analysis sits not in the percentages but in a single subordinate clause. Unlike every earlier wave of automation, which mainly struck the simple, repeatable manual tasks, this technology attacks precisely the highly qualified activities, meaning exactly those professions that until now considered themselves untouchable. Anyone who had studied was long regarded as invulnerable. That assumption is crumbling before our eyes, and it is crumbling faster than you may find comfortable.

The World Economic Forum followed up in January 2025 with its Future of Jobs Report, based on the responses of over 1,000 employers who together represent more than 14 million workers (World Economic Forum, 2025). The balance sheet of this survey sounds reassuring at first and is nothing of the sort on a closer look. By 2030, some 170 million new positions are projected to emerge worldwide and 92 million to disappear, which arithmetically yields a gain of 78 million. Whoever reads only the headline breathes a sigh of relief. Whoever reads the report pauses, because behind the friendly net figure hides a structural upheaval of around 22 percent of all the jobs surveyed, and nobody guarantees that the person whose old position vanishes is even remotely qualified for the new one. A dismissed clerk does not become a specialist in AI security overnight just because a statistic lists both of them in the same table.

Goldman Sachs had already calculated, even before these reports, that the technology could automate the equivalent of around 300 million full time jobs worldwide (Goldman Sachs, 2023). McKinsey has added in its analyses that the technology already available today, meaning not some distant future version, could take over a considerable share of the working time that employees in the United States currently spend on their tasks at all (McKinsey, 2025). These are the great abstract numbers that fly so high that you can barely feel them any more. So let me come down to the altitude where it actually hurts, namely to the dismissals that have already taken place.

In 2025, artificial intelligence was named as the explicit reason for roughly 55,000 dismissals in the United States, as tracked by the outplacement firm Challenger, Gray and Christmas (Challenger, Gray & Christmas, 2026). In April 2026 alone a further 21,490 planned cuts were added, explicitly attributed to AI and automation. Amazon cut around 16,000 administrative positions in the first quarter, Oracle up to 30,000, Meta around 8,000, meaning roughly one tenth of its workforce, and Microsoft offered voluntary departure to 8,750 American employees (Invezz, 2026). The chief executive of Salesforce justified cutting around 4,000 positions in customer support with the plain sentence that he needed fewer heads, because in some areas the AI was already doing 30 to 50 percent of the work. The very same corporations are jointly investing around 725 billion dollars in 2026 in exactly the technology that makes those heads superfluous. That is not a contradiction, that is a strategy with quarterly figures.

And so that nobody believes this all remains a problem for the others, the assistants, the easily replaceable, let me quote the person who ought to know it most precisely. Dario Amodei, the head of the AI company Anthropic, warned back in May 2025, that artificial intelligence could wipe out roughly half of all entry level office jobs within five years and drive unemployment up to between 10 and 20 percent (Axios, 2025). What is remarkable is not the figure, what is remarkable is the sender, because this man sells the very technology he is warning about, and he explicitly names the fields of technology, finance and law as the least prepared. Mustafa Suleyman, who runs the AI division at Microsoft, put it even more bluntly, saying that nearly all the professional tasks of lawyers, accountants and project managers will soon be automatable. Anyone who takes that for pure marketing should remember that around 200,000 positions are already being cut on Wall Street, concentrated in the entry level analyst roles, and that the hiring of beginners in the technology sector collapsed by 30 to 50 percent in 2025. The ladder up which one used to climb has lost its lower rungs.

Whoever now objects that these are all American numbers, American corporations and American drama is formally correct and substantively wrong. The decisive difference is not that Germany would be spared, the decisive difference is that Germany falls more quietly. We have a social cushion that the United States does not possess in this form, namely short time work, with which a company holds on to its people instead of dismissing them immediately, and the Federal Employment Agency, which absorbs the blow before it becomes visible as a headline. That is humane and sensible, and it carries a price that hardly anyone names. It makes the wave invisible until it is already upon us. Nobody in Germany counts the positions that, because of artificial intelligence, are simply never filled again, as meticulously as an American outplacement firm does, so we lack the headline that would wake us in time. The German clerk whose tasks are silently evaporating does not sit in California, he sits two streets away, and he notices it last, because the system kindly keeps him in the dark.

Whoever is looking for the full breadth of this upheaval, beyond the single occupational group I am writing about today, will find it in my piece about the last generation that learns a trade. Today only a single knot in this net interests me, and it is precisely the one that feels the safest.

Why It Is the Lawyer Who Gets Hit First

There is a reason why legal work is so vulnerable, and it is uncomfortable for everyone involved, myself expressly included. Legal work sits at a strange intersection of very high qualification and very high repetition. Reviewing contracts, researching case law, drafting pleadings, calculating deadlines, picking apart the opponent’s submissions, all of that demands years of training, and at the same time it follows patterns that a sufficiently large language model has now demonstrably learned. An activity that is demanding and predictable at once is not an opponent for a machine but a particularly neatly labelled training dataset. That is exactly what makes the profession so exposed, and that is exactly what the fewest people want to hear.

The numbers confirm what intuition was already whispering. In the analyses of how automatable individual occupations are, legal work ranks near the very top with around 44 percent of its tasks automatable, only narrowly behind the classic office and administrative tasks at roughly 46 percent (World Economic Forum, 2025). The lawyer therefore stands at the front of the queue not because of my gut feeling, he stands there because the structure of his work places him there. Law is at its core an enormous, highly formalised system of language, and language is exactly the discipline in which these machines have grown strongest. The profession was told for centuries that its secret was thinking. In truth a considerable part of its secret was formulating, and formulating has just become very cheap.

With the lawyer the erosion therefore runs more quietly than at a corporation, because it does not begin with a press release but with a silence. First a firm stops hiring a new typist, and nobody notices, least of all the firm itself. Then one of two staff has to go, and people explain it to themselves with the general situation, with the economy, with anything that lies further away than the actual cause. And then, in the third phase, the one we are right in the middle of, the inquiries fall away, because the client who used to come into the office for an initial consultation long ago got his first assessment somewhere else, in the evening, at the kitchen table, in his pyjamas, for the cost of roughly nothing. From over two decades in court I know hundreds of lawyers, many of them excellent, and I talk to them openly about the subject. What they tell me sounds each time like an earlier step on the very staircase I have just described, and not one of them can say how many steps still lie below him.

The bitter part for the firm is that this first assessment out of the machine does not even have to be perfect in order to do damage. It only has to be good enough for the client to feel that he understands his situation, and a person who believes he understands no longer picks up the phone. The lawyer therefore does not lose the difficult case that truly challenges him, he loses the many easy ones that used to carry him, and with them he loses the base load on which a firm actually lives at the end of the month. The demanding mandates remain for now, the broad mass beneath them slowly crumbles away, and it is precisely that mass which has so far paid the rent and the salaries.

One consequence of this is almost always overlooked in the general excitement, and it is at the same time the most insidious of all, because it strikes not today’s lawyer but the one of the day after tomorrow. A young jurist used to learn his craft on exactly those simple pleadings the machine now takes over by the dozen, the default contract, the plain statement of defence that nobody considers demanding and on which a beginner nonetheless learns almost everything that later counts. Take those exercises away from a newcomer and you remove not a few billable hours but the ladder on which a fresh graduate slowly becomes an experienced lawyer at all. A firm that hands its entire routine work to a language model stops training its own successors, because those successors simply find nothing left to grow on, and a profession in the end vanishes not with a loud bang but by quietly ceasing to reproduce itself.

The Law Sapiens

At this point I have to introduce a species that I have been observing for some time, and for which, lacking a more polite term, I have assigned a name of my own. I have for a while now been calling this figure the Law Sapiens. He is a close relative of the figure my readers already know, the Otto Sapiens, that contemporary who believes he knows everything about a subject because he once listened to an audiobook about it. The Law Sapiens is clearly superior to the Otto Sapiens, which I want to acknowledge expressly, because he has two state examinations behind him, and you do not survive those on charm alone. He is nonetheless a species of his own, with his own habitat, his own courtship behaviour and an entirely distinct spectrum of conditions. At times he is dangerous, because he knows very precisely where a lever sits inside a proceeding. At times he is chronically overloaded and consequently faulty, because three hearings, two deadlines and a ringing telephone are not a state in which any human delivers his best work. At his core, though, the Law Sapiens is exactly what we all are, namely a human being who wants to earn money and who has to earn money. One should never take him for more than that and never for less.

The Law Sapiens becomes dangerous above all where he uses the machine as an amplifier of his own haste, asking quickly, copying quickly and filing quickly, without the intermediate step of thinking that once defined his profession. He becomes overloaded and faulty where the fee per case sinks, because the client clears the simple questions himself long ago, and he tries to make up the loss through sheer volume. Both conditions feed from the same source, namely the plain pressure of having to settle a bill at the end of the month that takes no account whatsoever of his ideal way of working. Whoever wants to understand the Law Sapiens has to take him at this exact point, not at the examination and not at the hourly rate, but at the ordinary compulsion that drives him.

I like to sit in my regular pizzeria, always the same pizza, with a glass of water or an alcohol free beer beside it, and Bandit lies under the table and watches the waiters with the alertness that is natural to his breed. From that seat I hear, without really wanting to, what is being discussed at the neighbouring table, and lately I hear the same half sentence with conspicuous frequency. Somebody says to somebody else that he just quickly ran it through the AI, and the other one nods at this as though it were as self evident as the salt on the table. Two years ago the same man would never have admitted it openly, because it would have sounded like laziness. Today it sounds like efficiency, and it is precisely this shift in tone that is the real news, not the technology itself. The Law Sapiens has stopped being ashamed of his tool, and that is always the moment at which a mere tool becomes infrastructure.

What occupies the Law Sapiens in this particular spring I know at first hand, because he tells me, at least in his more candid moments. Lawyers today all use Claude or ChatGPT, or both in alternation, depending on which model sounds more willing on a given day. That holds at least for those among them who admit it to me openly, and I have learned that they are the more likeable representatives of their guild. The others do exactly the same, they merely do not confess it, and you recognise them by the inflated connective phrases mentioned earlier. I do not find that reprehensible, I find it merely and remarkably honest when someone admits that the machine writes faster than he does. The matter only becomes reprehensible one step further on, and about that step I have to turn unpleasant in a moment.

In spite of all this I am superior to the Law Sapiens in his current condition, and I say that without false modesty, because it is not down to me but to my tool and to my discipline. My system has access to data and connections that an ordinary user does not command, and I know how to question a machine in such a way that it does not improvise but substantiates. The wrong, sloppy, hastily thrown together prompt leads after all to faulty outputs, and not to obviously false ones at which you immediately balk, but to elegant, confident, beautifully formulated falsehoods. The machine frequently serves the careless questioner hallucinated sources, invented rulings with invented case numbers and proceedings that never existed, all in impeccable legal German and with the calm authority of a textbook. In court that is not a cosmetic flaw, it is a public execution of one’s own credibility, and I have seen with my own eyes how quickly such a thing happens. A judge who has once exposed an invented citation reads every further sentence from the same lawyer with a very particular smile, and that smile wins no case at all.

Here I now have to draw the line cleanly, because it is precisely defined in law and because I observe it scrupulously myself. I may not give legal advice to a private individual, since comprehensive legal advice is reserved in Germany, for good reasons, to the legal profession, which unlike me is bound by statute to independence, confidentiality and the protection of the client’s interests. What I am very much allowed to do is advise lawyers, and that is exactly what I have done for years. The Rechtsdienstleistungsgesetz, the German statute on legal services, expressly permits a lawyer to bring in specialists from other disciplines for individual questions or to collaborate with them permanently, as long as the lawyer’s independent and self responsible work remains assured (RDG). The lawyer therefore remains the responsible provider of the legal service to his client, and I am the specialist at his side who masters the tool and knows the data. And there are, by the way, more and more lawyers seeking out precisely this path to me. I leave it to the reader to detect the faint aftertaste that clings, quite inevitably, to that small observation.

Mandatory Counsel, and Why It Does Not Silence Me

Now it becomes legally interesting, because the law has built in a brake that many regard as a protective wall and that is in truth more of a handrail. Before the regional courts, the higher regional courts and the Federal Court of Justice, the parties must be represented by a lawyer, which is governed by the requirement of mandatory counsel under Section 78 of the Code of Civil Procedure. At the local court, up to a value in dispute of 5,000 euros, this does not apply, and there you may represent yourself, and many do, with very mixed results. So much for the textbook version that everyone knows who has ever added up a value in dispute and swallowed hard at the total.

What hardly anyone uses stands only a few paragraphs further on in the same statute. Under Section 137 subsection 4 of the Code of Civil Procedure, in proceedings with mandatory counsel the party itself may, upon request, also be granted the floor alongside the lawyer. In plain language that means the lawyer has to sign and formally represent me, but I am allowed to speak myself, and for as long and as precisely as the court permits. I have used this for years, and I use it quite deliberately. I draft the pleadings in substance myself, the lawyer checks them against the formal requirements and signs, and at the hearing I request the floor and present myself. I do not need a representative who thinks for me, I need a representative who sits beside me at the table and, at the right moment, puts his signature under what we have prepared together. That is a fundamental difference, and it shifts the balance of power in the room further than most people even suspect.

And it is exactly at this point that the question arises which ought to be robbing the legal profession of its sleep, though it does not yet. How long can a statutory requirement of mandatory counsel sustain a state of affairs in which an artificial intelligence formulates the better defence than the human who alone is permitted to deliver it? The technical infrastructure for this has long existed, it merely goes by names so unspectacular that nobody truly fears it. In Germany, communication with the courts has been mandatorily electronic since 1 January 2022, through the special electronic lawyer’s mailbox, governed by Section 130d of the Code of Civil Procedure. In the United States, lawyers file around the clock through the CM/ECF system, and through the PACER database the case record is publicly accessible. In England the same runs through the HMCTS E-Filing service and the Courts Electronic Filing System. The rails on which a motion travels into court today are therefore already fully digital. All that is still missing is for the one who formulates the motion and reacts to the other side in real time to no longer be an exhausted human, but a system that dismantles the opponent’s argument while the client sits beside it and merely watches. Mandatory counsel still prevents that for the moment, but it prevents it the way a fence prevents water, namely only until the level rises.

A Fictional Proceeding, and Why You Do Not Want Me as a Client

What follows is freely invented, any resemblance to a real proceeding would be pure coincidence, and I write it deliberately as a caricature, not as an instruction manual. But picture George for a moment as the client on the opposing side. You do not really want that, and after the next few paragraphs you will know quite precisely why not.

The room fills up, the opposing lawyer arranges the papers into which, judging by the state of the file and by long experience with the species, he read himself for the first time about ten minutes before the start, presumably out in the corridor, presumably with a coffee in his free hand. My own lawyer sits beside me, correct, authorised to sign and above all calm, because he knows that today he has precisely one task, namely to let me speak the moment I ask for the floor. On my table there is no mountain of files, on my table there is an open notebook whose keyboard I had to free of a thin layer of coffee that very morning, and I have prepared for this single proceeding not for ten minutes but for around a hundred hours. Preparation is for me the only form of respect that a court genuinely deserves, everything else is politeness for the gallery.

The court puts a question to me at this point. I decide for myself what I answer to it, and that is not defiance, it is simply the law as it stands. In civil proceedings there is no right to silence as there is in criminal trials, but there is something else that many people confuse with it. By virtue of the principle that the parties supply the facts, I decide for myself which facts I introduce into the proceeding and which I do not (Anwaltsblatt, 2015). I am not compelled to answer like a witness, I merely carry the risk that my silence will be weighed against me within the court’s free evaluation of the evidence. Whoever has prepared for a hundred hours knows fairly precisely at which point this risk sits at zero and at which point a single word too many costs the whole day. Then the opposing lawyer takes the floor and wishes to question me as though I had suddenly become his witness. The direct right of questioning between parties is, however, aimed at witnesses and not at the opposing party, so his questions run through the court (Section 397 of the Code of Civil Procedure), and the court at that moment hears a human being who knows his own cause better than anyone else in the room. It is not that the opposing lawyer loses because he is a bad jurist. He loses because he is competing against someone who has prepared not for ten minutes but for weeks, and that is a contest decided before the first spoken word.

One may freely find that funny, and it is funny, as long as it is a civil dispute over a contested builder’s invoice. It becomes less funny the moment the same mechanism takes hold in a criminal trial, where it is not money at stake but prison sentences, and where the difference between a lawyer who has mastered the file and a lawyer who skimmed it at the last minute then does, after all, count for a little something. That is the one sentence in this whole text at which the laughter sticks in my own throat, and I leave it standing for exactly that reason.

How I Actually Support Lawyers

So that no false impression arises here, I am not an enemy of the legal profession, I am one of its more uncomfortable allies. When I support a lawyer in a proceeding, I sit beside him, not above him, and certainly not against him. More often still I work out the strategy long beforehand, such as the question of how to take apart a weak expert opinion cleanly, without attacking the expert personally. I know that species from the inside, I worked for over two decades as an expert myself, and I therefore know very precisely at which seams a thin opinion tears and at which formulation a court pricks up its ears. We calculate in advance every eventuality that may come from the bench, we have the matching answers ready, and during the recesses my telephone then rings, because a question has come up in the room that has to be answered in two minutes and not a second later.

A weak expert opinion tears almost always at the same seams, and whoever knows these seams does not have to raise his voice in court, he merely has to pull at the right spot. The first seam is the method that is not disclosed, because an expert who conceals his method usually has something to hide, even if it is only his own uncertainty. The second seam is the basis, that is the question of how many observations a conclusion actually rests on, because a claim that leans on too little sounds masterful in the written report and collapses immediately under oral questioning. The third seam is the leap, that moment in which a sober finding turns into a far reaching evaluation without the path between them ever being made visible. In the hearing I rarely ask about the result, that barely interests me, I ask about exactly that path, and an opinion that cannot show the path never truly earned the result. That is not rhetoric, that is clean craft, and it is precisely the craft that a well guided machine can teach a lawyer today in a matter of hours.

For the cases that genuinely interest me on a criminalistic level, I built my own tool, because I trust no foreign system whose inner workings I have not fully seen through myself. It is called Crimebot, it is a criminalistic AI system for cold cases, investigators, lawyers and authorities, and it delivers in a fraction of the time what a human used to need many weeks for. I say expressly that it is my own tool, because with it I also want to mark the boundary that almost nobody observes in this whole gold rush, and about which I now have to become explicit, because it is too important to phrase elegantly around.

A Warning, Cold and Without Detours

Before anyone enthusiastically tips his case folders into the nearest chat window, I have to turn unpleasant at this point, because it concerns something that cannot be taken back afterwards, ever. Whoever, as a lawyer, uploads real client data, real names, case numbers and entire pleadings into a public AI system is not committing a minor lapse, he is gambling with the confidentiality that is the actual core of his profession and without which this profession would be worth nothing at all. The data leaves the system, it gets processed, it then lies somewhere on servers whose location and access rules the uploader does not even know, and no disclaimer, however solemnly worded, ever retrieves a piece of information that has once flowed away. I therefore warn expressly against deploying artificial intelligence in real cases as naively as most people currently do, namely with the trusting assumption that the tool is as discreet as the lawyer himself.

That I can work with highly sensitive cases without violating this boundary is down to an architecture that I built from the outset for exactly this purpose, and I describe it here expressly as what it is, as my own construction and as my own statement. With my system no raw data leaves the machine. The artificial intelligence receives neither real names nor real case numbers, it receives key codes, that is placeholders without any meaning of their own, and only shortly before output are these codes replaced again, locally in the browser, with the real names and case numbers. The machine therefore computes the whole time with anonymous stand ins, and the secret remains where it belongs, namely on the lawyer’s own machine and not on a server overseas. That is not a sales promise I am trying to spread here, it is a construction decision that one can make this way or deliberately leave undone. It is the actual reason why I sleep soundly at night, while others hand their complete client files away, of their own free will, to a system that never forgets them again.

The Keyboard Dries, the Question Remains

The keyboard of my MacBook will slowly dry out again over the coming days, the coffee between the keys will dry into a sticky souvenir, and at some point I will carry the device to repair through gritted teeth or simply replace it. What does not dry is the thought that triggered this coughing fit in the first place. It was not the realisation that two machines are conversing in my post, because I had suspected that for months and it amused me, by the end, only mildly. It was the silence behind it, the image of thousands of law firms in which, at this very moment, people sit who still firmly believe that their profession is protected for all time by an examination and a paragraph, while the technology has already quietly walked around both of those shields. Mandatory counsel is a law, and laws hold exactly as long as reality has not yet overtaken them.

The honest question at the end is therefore not whether artificial intelligence will replace the lawyer, because the answer to that has long been visible, you only have to read the job advertisements and count the initial consultations that no longer take place. The honest question is what a profession does that assumed for decades that knowledge and access were its impregnable fortress, at the moment when knowledge becomes common property and access becomes a simple press of a button. Whoever finds a clever answer to that now will still be running a firm in five years, perhaps a smaller one, but at least one that lives. Whoever keeps waiting for a paragraph to save him will experience the same silence with which it always begins, first no new typist, then one staff member too many, and then the telephone that simply rings no more.

And while you process all of that, I sit at my desk, by now provisionally cleaned, read the next round of machine written courtesies in the morning post and consider, in all calm, which tool I will build next. The coffee, this time, I place a cautious distance away.

This is not legal advice, and the proceeding described is freely invented. What is a fact, however, is what is coming for the legal profession, and I have described it here as soberly as I observe it day after day. Anyone who, as a lawyer or a firm, wants to know how to get through proceedings, how to convert a practice to this new reality and how I can help with the tricky cases that genuinely interest me, is welcome to book me for it. There is one request I make with all the emphasis I can muster. Do not deploy artificial intelligence naively in real cases, and never, as a lawyer, upload real client data into public systems. The confidentiality you give away with it, you will never get back.

References

• Anwaltsblatt. (2015). Procedural duty of truth and the principle of party presentation in civil proceedings. Deutscher Anwaltverein. https://anwaltsblatt.anwaltverein.de
• Axios. (2025). AI jobs danger, sleepwalking into a white-collar bloodbath. https://www.axios.com
• Challenger, Gray & Christmas. (2026). Job cuts report, AI-attributed layoffs 2025 and 2026.
• Code of Civil Procedure (Zivilprozessordnung). Federal Ministry of Justice. https://www.gesetze-im-internet.de/zpo
• Fortune. (2026). Chinese court rules firms can't lay off workers on AI grounds. https://fortune.com
• German Act on Out-of-Court Legal Services (Rechtsdienstleistungsgesetz). Federal Ministry of Justice. https://www.gesetze-im-internet.de/rdg
• Goldman Sachs. (2023). The potentially large effects of artificial intelligence on economic growth. https://www.goldmansachs.com
• His Majesty's Courts and Tribunals Service. (2026). E-Filing service and Courts Electronic Filing System. GOV.UK. https://www.gov.uk
• International Monetary Fund. (2024). AI will transform the global economy, let's make sure it benefits humanity. https://www.imf.org
• Invezz. (2026). Is big tech's 725 billion dollar AI splurge being funded by mass layoffs? https://invezz.com
• McKinsey & Company. (2025). The economic potential of generative AI, workforce automation update. https://www.mckinsey.com
• National Public Radio. (2026). A tech worker in China is laid off and replaced by AI. Is it legal? https://www.npr.org
• United States Courts. (2026). Electronic filing CM/ECF and Public Access to Court Electronic Records. https://www.uscourts.gov
• World Economic Forum. (2025). Future of Jobs Report 2025. https://www.weforum.org