Imagine a court where none of the documents in the case were written by a human being. The initial petition was generated by one language model. The answer to the complaint, by another. The legal opinion supporting the argument came from a third algorithm, trained to sound like a great jurist. And the judgment? It was drafted by a fourth machine, installed in the judge’s office, tasked with reading what no human had time to read. The question remains: where, exactly, is justice in this scenario? And where is the person it is supposed to protect?
This scenario is not a futuristic hypothesis. It is the present, and it has already reached the courts. Artificial intelligence is no longer just a research tool; it has taken over the work that, for centuries, has defined the legal profession: arguing cases in writing. Law firms are drafting appeals at scale using increasingly refined prompts. Some are already tailoring these prompts not to convince the judge, but to convince the AI that screens the appeals before any human even looks at them. On the other side, the courts are responding in kind, using algorithms to summarize, classify, and, often, draft the decisions. A closed and somewhat absurd loop has formed: a machine writes to deceive the machine that will render the judgment. The human, when they appear, is there only to sign.
Proof that this system works came from the United Kingdom. In June 2026, a firm that bills itself as entirely AI-based—Garfield AI—won its first court case. The platform handled every aspect of a debt collection case: the initial letters, filing the lawsuit, and witness statements. The client reportedly paid around four hundred pounds to try to recover seven thousand, while the opposing party went to court with a solicitor and a barrister, following the traditional and expensive model. The AI handled the behind-the-scenes work that has always been the responsibility of human lawyers. And it won. The key detail relevant to this argument is what the machine did not do: the oral argument, during the approximately three-hour hearing, was delivered by a flesh-and-blood barrister. It was a human, standing before the judge, who presented the defense at the one moment when physical presence still mattered.
It is worth recalling how the Brazilian legal process has reached this point, because the vulnerability is structural and has a history. The national tradition has always been that of written proceedings—formerly on paper and today in stacks of PDFs. A taste has been cultivated for lengthy briefs, abundant citations, and the transcription of legal doctrine and case law that repeats itself over dozens of pages. There was a time—and it wasn’t that long ago—when litigants would commission legal opinions from renowned jurists to support their arguments, and the name of the author carried almost as much weight as the argument itself. That was real symbolic capital. Well, today, in the concrete world of the courts, these opinions are often not even read. Not out of intellectual contempt, but for practical reasons. A judge drowning in thousands of cases has no way to leisurely read a brief of a hundred-odd pages, no matter how illustrious the signature at the bottom may be. The legal opinion becomes an attachment. The appendix becomes a number. And what was once high-level legal debate is reduced to dead volume within a system that measures productivity by the number of judgments handed down. If sophisticated written work was already not being read when it was human-generated, it is naive to suppose that it will survive once it is mass-produced by machines.
This leads to the structural explanation of the problem. Generative AI excels precisely at what the Brazilian legal process values most: drafting, compiling, citing, and organizing persuasive text. If the essence of legal disputes continues to be a battle of documents, that battle will be won by whoever has the best model—not by whoever is right or has talent. The consequence is direct and unsettling. There will no longer be a need for so many litigation attorneys, because the intellectual and repetitive aspects of litigation are precisely what will be automated first. The role of the lawyer as a drafter of legal documents is doomed. And the Constitution, which requires trials to be conducted by robed judges, is being quietly circumvented in practice, as the deliberation and drafting of decisions shift to office algorithms that no one has elected or invested with jurisdiction.
There is a way out, and it does not involve banning the technology, which would be futile. It involves changing the framework. The solution is to shift the focus of the process from the written word to the spoken word, in a model that resembles the Anglo-Saxon system. In this model, written submissions—motions and requests—are short, direct, almost like a form that merely defines the subject matter of the dispute. The actual debate, the presentation of evidence, and the persuasion of the judge all take place live, in the courtroom, with both parties in the presence of the judge, who listens and decides on the spot. When the weight of the argument shifts from the document to the voice, the lawyer once again becomes irreplaceable. A machine can draft a flawless contract in seconds, but it cannot cross-examine a witness by picking up on hesitation in their voice, cannot read the discomfort on the judge’s face, and cannot adjust its strategy mid-hearing because the opponent’s rebuttal was better than anticipated. Improvisation, intuition, reading people, and the ability to react in real time remain biological attributes, and it is precisely through these that oral advocacy restores the value of legal work.
The institutional implications of this shift are significant, and they should not be downplayed. Restoring oral proceedings means bringing humanity back to the trial, requiring the judge to engage directly with the parties rather than delegating the reading and drafting of documents to staff and software. It also means speed and transparency, because ending the endless back-and-forth of petitions, replies, and counter-replies eliminates years of litigation. But it comes at a high price. It requires profound legislative reform—likely a new code of civil procedure—and something even more difficult to achieve: a cultural shift in law schools, the legal profession, and the judiciary. Judges would have to leave the comfort of their chambers and assume an active, visible role in court hearings. Lawyers would have to trade the security of copy-and-paste for the discipline of public speaking, quick thinking, and negotiation. AI would remain present—and welcome—to fill out forms, organize facts, and handle the behind-the-scenes legwork. It just wouldn’t decide cases by proxy anymore.
There is an even deeper layer, one that goes beyond the legal sphere. Yuval Harari argues that humans came to dominate the planet not through physical strength or individual intelligence, but through the ability to cooperate on a large scale around shared fictions—realities that exist only because we collectively believe in them. Money, governments, and the judiciary itself have no material existence; they are invented institutions that function as long as collective belief sustains them. A court has authority only because an entire people agrees to pretend, every day, that that courtroom and that judge’s robe carry real power. When the act of judging is transferred to a machine, we are not merely replacing a worker with cheaper software. We are entrusting to the algorithm the guardianship of one of the founding fictions of human cooperation. And it is difficult to collectively believe in a justice system that no one, from the inside, experiences as human. Losing the power to judge to the machine is, in this sense, giving up part of what has made us dominant in the world, because it means giving up our ability to uphold, together, the institutions we invented to live in society.
The conclusion is unsettling either way. If the current electronic written process remains in place, mass litigation will be absorbed by the cold efficiency of language models, and citizens will lose the only guarantee that truly matters—that their dispute was considered by someone capable of empathy and fairness, rather than merely calculated by a system optimized to meet targets. Artificial intelligence does not kill the legal profession. It kills the lawyer who has reduced himself to a mere scribe. What it puts to the test is whether the Brazilian justice system will have the courage to do what it has been avoiding for decades: to stop writing so much and return to speaking—eye to eye, out loud—before a judge who listens. Perhaps the ultimate paradox is this: it took a machine learning to write almost like a human to remind us that judging, in the end, is still a human endeavor.
More about
Politics
