Where do we, humans, stand?

The first AI courtroom controversy was about transparency. The real disruption begins later, when machines are trusted enough to argue, decide, and appeal, and humans must redefine what professional value and justice actually mean.

The artificial lawyer had barely started speaking when the judge interrupted.

“Hold on.
Is that counsel
for the case?”

It was not.

The man appearing before the court had submitted a prerecorded argument using an AI-generated avatar. He had not clearly informed the court that the person delivering the argument was synthetic.

The judge was not impressed.

“I don’t appreciate
being misled,”

“I don’t appreciate being misled,” she said, before ordering him to switch the video off and present the argument himself.

At first sight, the scene appears to capture the central conflict of the AI era: machines entering human institutions before those institutions are ready to accept them.

But the controversy was not really about whether the artificial speaker could produce a valid legal argument. It was about disclosure, identity, authorization, and the rules of the courtroom.

Those questions matter now because we are still in the first phase of AI adoption.

We are still asking whether people know when they are interacting with a machine. We are still debating transparency, trust, accountability, and human oversight.

But this is only the beginning.

The harder moment will arrive after trust has been established.

After AI systems become reliable enough to be accepted. After citizens use them successfully. After they are certified, audited, insured, and legally authorized to operate.

What happens when the AI is no longer pretending to be a lawyer, but is officially permitted to become one?

And what happens when the judge is an AI too?

Today, the responsible position seems obvious: organizations should disclose when AI is involved.

People should know whether they are speaking to a person, an avatar, a chatbot, or an autonomous system. They should know where the machine’s role begins and where human responsibility remains.

Without that transparency, trust breaks quickly.

But disclosure does not resolve the deeper economic and institutional conflict.

Imagine that an AI legal agent clearly identifies itself before every interaction. It explains its limitations. It operates under defined regulations. Its performance has been independently evaluated.

Now imagine that it can read every relevant document, retrieve legal precedents instantly, identify inconsistencies in testimony, model the strategy of the opposing side, communicate in dozens of languages, and serve thousands of clients simultaneously.

It is available at any hour.

It does not become tired. (certain token limitations may apply, but ok)

It does not forget a case. (certain memory buffer may apply, but ok)

It does not charge by the hour. (token price is gonna be a b!7ch, but ok)

At that point, the main question is no longer whether citizens trust the system.

The question becomes what happens to will do the professionals whose income, status, and identity were built around being the people capable of delivering that value.

This is the real AI shakedown. It feels multiple stakeholders will need to take action.

Most professionals are comfortable with AI as long as it remains a tool.

A lawyer can use AI to research cases, summarize documents, review contracts, or prepare an initial draft. The lawyer remains at the center of the relationship.

The hierarchy is preserved.

The lawyer leads.

The machine assists.

The lawyer interprets. (soon machines will do better than humans, but ok)

The lawyer signs. (to be seen, when machines are “certified” in knowledge and defined criteria, we will see, but ok)

The lawyer gets paid. (as soon as machines can get paid and earn for deliverables we will see, in practice the owners of those agents will get paid, but ok)

An autonomous legal agent changes this structure. (it is already happening, but ok)

It does not help the lawyer provide the service. It competes to provide the service directly.

This distinction matters because AI does not compete like another human professional.

A highly skilled lawyer has limited time. The lawyer can only attend a certain number of clients, review a certain number of cases, and appear in one courtroom at a time. (The attention economy creates multiple dimensions no bound by conventional physical and human laws, but ok)

An AI agent is replicable.

One highly capable system can become one thousand, one million, or ten million active instances.

It can serve a global market without producing ten million new professionals.

That is not simply an increase in productivity. It is a structural change in the supply of professional intelligence.

And when something previously scarce becomes abundant, its economic value changes. (The whole dynamic changes, initially who ons those new powerful assets will be in advantage. The question is, will this be similar to the impact of AGI? or will it eventually be something that everyone will have access to?)

This is where the conflict becomes uncomfortable.

Would lawyers support a system
capable of independently
representing citizens?

Some would and many probably would not. Who knows.

They would raise serious concerns about confidentiality, liability, hallucinated precedents, manipulation, conflicts of interest, ethical duties, and the right to competent representation.

These concerns would be legitimate.

They would also protect the profession from competition.

Both things can be true at the same time.

Professional regulation has always had two functions. It protects society from unqualified practitioners, but it also defines who is allowed to enter a market and sell expertise.

AI will force us to examine where public protection ends and professional protectionism begins.

Lawyers may argue that citizens deserve a qualified human representative. The best representative possible.

But many citizens cannot afford one today.

For them, the real choice is not between an excellent human lawyer and an artificial one.

It is between an artificial lawyer and no meaningful legal support at all.

A regulated AI agent may be worse than the best attorney in the country and still be better than facing a complex legal process alone.

Preventing its use could protect vulnerable citizens from a defective machine.

It could also protect an inaccessible legal system from a cheaper competitor.

This tension will appear across many professions.

Doctors will defend patient safety.

Teachers will defend educational quality.

Accountants will defend financial integrity.

Consultants will defend strategic judgment.

In many cases, they will be correct.

They will also be defending their position in the delivery of value.

The usual optimistic argument is that AI will remove repetitive work and allow humans to move into more strategic roles.

There is truth in this.

The lawyer of the future may spend less time searching for documents and more time defining strategy, negotiating, understanding human motivations, managing exceptional cases, and challenging the recommendations of automated systems.

But this argument hides a major mathematical problem.

Not every lawyer can become a senior strategist.

And the economy will not need as many strategists as it once needed people performing execution.

If one lawyer supported by advanced AI can do the work that previously required ten professionals, we do not automatically obtain ten more strategic lawyers.

We may obtain one senior lawyer, one technical supervisor, and a collection of automated agents.

The other positions may simply disappear?

This is especially disruptive because professional careers are built as pyramids.

Junior lawyers perform research, prepare documents, review evidence, and learn by working through thousands of smaller tasks. Over time, some of them acquire the experience required to become senior lawyers, partners, judges, or respected legal strategists.

But AI is most capable of absorbing precisely this entry-level work.

If the lower levels of the pyramid disappear, how will future experts develop?

A profession can automate its junior tasks so successfully that it destroys its own training system. Though we will see if this holds true, mainly because we see growth in hiring entry positions especially in consulting firms where AI has been told that it would replace may of those positions.

AI may democratize legal access for citizens while making the legal profession more concentrated and elitist.

Fewer people may enter.

Fewer may progress.

Those who already possess reputation, relationships, capital, and institutional authority may capture most of the remaining human value.

Keeping a human judge appears to solve the legitimacy problem. Will judges use or be supported by Ai Judges Copilots to be more precise and knowledge in terms of context, motivations, and other details in order to deliver the best outcome possible?

The AI lawyer may research, argue, and respond, but the final decision remains with a human being.

The judge becomes the institutional anchor: the person who understands context, recognizes ambiguity, weighs competing interests, and accepts responsibility.

But this arrangement may also be temporary.

If AI can process evidence faster, compare thousands of similar cases, identify contradictions, apply procedural rules consistently, and reduce court backlogs, the same economic and performance pressure affecting lawyers will eventually reach judges.

The transition will probably happen gradually.

First, AI summarizes filings.

Then it identifies relevant law.

Then it drafts routine decisions.

Then it recommends sentences, settlements, or outcomes.

Then the human judge reviews and approves the recommendation.

Eventually, the institution may discover that human review has become largely ceremonial.

The machine analyzes the case.

The machine proposes the outcome.

The human signs.

At that point, who is really making the decision?

And if the automated system produces more consistent decisions, fewer procedural errors, and faster outcomes than human judges, another uncomfortable question appears:

Is it ethical to insist that judgment must remain human only human?

This is where the debate moves beyond human versus machine.

The next stage is agent to agent: A2A.

A citizen’s legal agent prepares and files a claim.

The institution’s agent reviews it.

A defense agent challenges the evidence.

A judicial agent evaluates the arguments, asks questions, applies the law, and produces a decision.

An auditing agent examines the process for bias or procedural errors.

An appellate agent decides whether the original judgment should stand.

The entire exchange may happen at machine speed.

The agents could analyze thousands of precedents, simulate multiple legal strategies, test alternative interpretations, and negotiate a settlement before a human lawyer finishes reading the first brief.

From the perspective of efficiency, this could be extraordinary.

From the perspective of citizenship, it could be deeply alienating.

The person affected by the decision remains human-ish.

The person loses the home, the job, the benefit, the custody case, the immigration status, or the freedom.

But the real argument may take place between systems communicating in ways the citizen cannot understand, at a speed the citizen cannot follow, using assumptions the citizen cannot see.

A2A justice could dramatically reduce the cost and latency of the legal system.

It could also transform citizens into spectators of their own cases.

An A2A legal system creates risks that go beyond the errors of one isolated model.

Suppose the legal agent, the defense agent, the judicial agent, and the appellate agent are built on the same foundation model.

Or perhaps they come from different companies but were trained using very similar data and design assumptions.

The system may appear adversarial.

One agent argues against another.

A third agent judges between them.

But they may still share the same blind spots.

They may interpret uncertainty in similar ways.

They may reproduce the same historical biases.

They may ignore the same forms of evidence.

They may all reach agreement not because the outcome is just, but because they are different interfaces built on the same technological worldview.

Adversarial institutions depend on disagreement.

Plaintiffs, defendants, lawyers, judges, and juries bring different incentives, experiences, values, and interpretations into the process.

Their conflict is inefficient, but it can reveal weaknesses that one centralized intelligence would miss.

If every institutional agent thinks in essentially the same way, we could automate the appearance of disagreement while removing its substance.

The future may therefore require an algorithmic separation of powers.

The system representing a citizen should not be controlled by the same organization that provides the judicial agent. (Governance)

The auditing model should not share all the assumptions of the model it evaluates. (Governance)

The appellate layer should have genuine independence, not simply a different prompt. (Governance)

Model diversity could become as important to automated justice as judicial independence is to human justice.

This brings us to the future role of humans. This is key and we should be thinking about it very strongly.

If machines become responsible for research, argument, prediction, and decision-making, humans will need to become much more capable in areas such as ethics, equity, fairness, governance, and institutional design.

These subjects are often described as soft skills.

They are not. If a a role can be defined, measured and manageable it changes everything?

They will become core infrastructure.

A machine can optimize a defined objective. But fairness is not one universal objective waiting to be discovered.

Should similar cases always receive identical treatment?

Should the system compensate for historical disadvantage?

Should it optimize overall accuracy or protect the group most exposed to error?

Is a false conviction worse than a false acquittal?

Should a system prioritize consistency, mercy, equality, proportionality, or social stability?

How much efficiency should society sacrifice to preserve a person’s right to be heard?

These are not engineering questions with purely technical answers.

They are political and moral conflicts.

The machine may calculate the consequences of each option.

Humans must decide which consequences are acceptable.

We will need people capable of translating social values into institutional constraints.

Not simply asking whether a model is accurate, but deciding what it is allowed to optimize.

Which errors are intolerable?

Who carries the risk?

Who benefits from the efficiency?

Who has the power to challenge the outcome?

And where must human judgment remain available, even when it is slower and more expensive?

Many institutions will respond by promising a human in the loop.

This sounds reassuring. Until it is not enough.

But a human presence does not automatically create human control.

Imagine a judge who receives hundreds of AI-generated recommendations every day. The system has already analyzed the facts, classified the case, selected the relevant precedents, and proposed a decision.

The judge has only a few minutes to review each recommendation.

Contradicting the system requires extra explanation, creates delays, and may reduce the judge’s performance score.

Technically, the human makes the final decision.

In practice, the machine controls the process.

The human becomes a liability shield.

The institution can claim that no automated system made the decision, even though the human had neither the time nor the information required to exercise independent judgment.

Real human oversight requires more than a signature.

The human must have authority, time, competence, access to evidence, and institutional protection when disagreeing with the model.

Otherwise, “human in the loop” is not governance. It is just a check list item amongst many.

It is pure theater.

In an automated institution, the right to appeal may become more important than the right to an initial human decision.

An AI system may process many routine cases more effectively than people. It may reduce waiting times, eliminate some forms of inconsistency, and make services available to citizens who are currently ignored.

Rejecting automation completely could preserve human dignity for some while maintaining delay and exclusion for many others.

But citizens must retain a meaningful way to challenge the machine.

Not simply receive an automated explanation.

Not simply click a button requesting review.

They need a process capable of bringing context back into the decision.

A place where exceptional circumstances can be heard.

A place where the system’s assumptions can be questioned.

A place where an institution accepts responsibility rather than pointing to an algorithm.

The appeal may become the moment when automated administration becomes human governance again.

The first generation of AI made information cheaper. The agentic generation will make action cheaper.

As machines learn to research, argue, negotiate, decide, and transact, professional value will shift from producing intellectual work to governing its consequences.

Lawyers may become less valuable for knowing the law and more valuable for defining strategy, accepting responsibility, and defending the human interest when automated systems fail. Judges may move from processing every case to protecting rights, exceptions, proportionality, and appeal.

This transition will not be smooth. Many professionals built their identity around expertise that society considered scarce. AI threatens not only their income, but the story they were told about why their work mattered.

Some will adapt. Some will use regulation to slow the competition. Some roles will become more valuable. Many will disappear.

And even the strategic layer may not remain exclusively human. “Humans will become more strategic” may describe the next stage, not the final one.

The judge in the viral video was right to stop an undisclosed artificial presenter. But that moment belongs to the first chapter of the story.

The next AI lawyer will identify itself. It may be authorized, audited, insured, and more effective than many human professionals. Then another AI will answer from the bench.

At that point, the question will no longer be whether machines are intelligent enough to participate in justice. It will be whether humans are capable of designing justice between machines.

A2A systems may make legal services faster, cheaper, and more accessible. They may also create institutions where citizens face decisions that are efficient, consistent, and almost impossible to contest.

The human role will not be to outperform every machine. It will be to decide what machines are allowed to optimize, where their authority ends, and when a person has the right to interrupt the process.