AI Has Finally Become a Manager

A group of 26 Meta employees has gone to court with a striking allegation: that internal AI systems, activity monitoring, productivity scores and measures of AI use helped identify people for layoffs — while failing to account properly for medical, parental and family leave.

The employees say the system treated periods of legally protected absence as reduced performance. People who were pregnant, ill, disabled, caring for relatives or bonding with a new child could not accumulate the same activity, output or AI-adoption scores as colleagues who were at work.

Meta denies the allegations. It says workforce and organisational decisions were made by people, not AI.

That distinction matters legally.

As reassurance, it is less impressive.

Because if the machines did not make the decisions, then human managers may have looked at measurements that could not understand illness, pregnancy, disability or caregiving — and decided the numbers were good enough.

So perhaps AI has not replaced management after all.

Perhaps it has merely given management a larger spreadsheet.

The machine measured exactly what it was told to measure

The allegations have not been proven. A judge declined an emergency request to stop the layoffs, while leaving open the possibility of reconsidering the decision if more evidence emerges about whether and how AI was used.

But the broader problem does not depend on the final result of one lawsuit.

Organisations are increasingly using automated tools to measure productivity, rank employees, analyse communications, track activity and recommend decisions. The sales pitch is familiar: more data, less bias, greater consistency and faster decisions.

Sometimes that is true.

Sometimes the system measures keystrokes because keystrokes are easy to count.

It measures logins because logins produce clean data.

It measures messages, documents, code, calls, tokens and visible activity because they fit nicely into a dashboard.

Then someone quietly makes the leap from measurable to valuable.

A person who sends more messages appears more engaged.

A person who produces more visible activity appears more productive.

A person who uses the approved new tool more often appears more adaptable.

A person who is not there appears to be doing very little.

This is mathematically correct.

It is also how a pregnant woman, a cancer patient, a disabled employee or a father on parental leave can become a low-performance data point without anyone ever instructing the system to discriminate.

The machine did not misunderstand the context.

The context was never included.

Objective is a beautiful word

Managers love the word “objective.”

It suggests fairness, discipline and freedom from personal prejudice. It also suggests that the decision arrived from somewhere else.

“The data shows.”

“The system identified.”

“The model ranked.”

“The dashboard indicates.”

Nobody says: “I chose to treat this metric as a reliable description of a human being.”

Numbers do not remove judgement. They hide where judgement entered the process.

Someone chose what to measure.

Someone decided which data mattered.

Someone assigned the weights.

Someone decided how absence would appear.

Someone chose the threshold.

Someone decided whether a manager would question the result or simply admire the professional-looking chart.

By the time the ranking reaches senior management, all those choices have hardened into something called evidence.

The output looks neutral because the arguments happened earlier, inside the design.

AI has mastered middle management

For years, people worried that artificial intelligence would become too human.

The more immediate danger is that it becomes middle management.

It has already learned the essential habits:

• measure what is easy rather than what matters;
• confuse visible activity with useful work;
• rank people without understanding what they actually do;
• remove inconvenient context from the decision;
• produce a colourful dashboard;
• call the result objective;
• leave accountability with somebody else.

This is not because AI is malicious.

It is because organisations often buy technology to automate management habits they were too lazy to question.

If the original performance system rewards noise, AI can reward noise faster.

If the organisation confuses availability with commitment, AI can monitor availability continuously.

If the company values adoption theatre, AI can count how often employees use the fashionable tool.

If leaders want a ranking that makes a painful decision look scientific, AI can provide one with several decimal places.

The technology adds speed, scale and authority.

It does not add wisdom unless wisdom was designed into the process.

The human in the loop may be decorative

Companies often respond to concerns about automated decisions by saying that a human remains in the loop.

Good.

What does the human do?

Do they have time to challenge the result?

Can they see which data created the score?

Do they understand the system’s limits?

Are they expected to use judgement, or expected to approve the recommendation unless they can prove it is wrong?

Will challenging the system delay an urgent restructuring?

Will the manager be praised for protecting context, or criticised for resisting data-driven decisions?

A human who clicks “approve” is technically in the loop.

So is a rubber stamp.

The phrase means little unless the human has information, authority, time and responsibility.

Otherwise the organisation gets the worst of both worlds: machine logic without machine accountability, and human authority without human judgement.

The productivity theatre problem

Many organisations still cannot define productivity properly.

They know it when they see a busy calendar.

They recognise it in a fast reply.

They feel reassured by a green status light, a crowded inbox and a large number of documents.

They struggle with the employee who prevents a problem quietly, protects an important relationship, notices a bad assumption, teaches a colleague, preserves institutional memory or makes one decision that saves three months of pointless work.

Those contributions are hard to count.

So they are often treated as secondary to the things that can be counted.

AI does not create this stupidity.

It industrialises it.

An automated system can turn weak assumptions into thousands of consistent decisions. Consistency sounds like fairness until the same mistake is applied to everyone.

At small scale, a poor manager misunderstands a few people.

At digital scale, an organisation can misunderstand the workforce by design.

Use AI as a signal, not a verdict

None of this means organisations must abandon AI in workforce management.

It means they must stop treating automation as a substitute for responsibility.

A defensible system would make several things unavoidable:

• protected leave and approved accommodations must be excluded or adjusted explicitly;
• employees must know what data is being used and for what purpose;
• managers must be able to see why a score was produced;
• automated rankings must trigger review, not determine the answer;
• people must have a practical route to correct bad data and challenge conclusions;
• the organisation must test whether apparently neutral measures harm particular groups;
• a named human decision-maker must remain accountable for the result.

This is slower than pressing a button.

It is also management.

The future arrived with plausible deniability

The Meta case will be decided on evidence, not sarcasm. The claims may be proved, rejected or narrowed significantly.

But the case describes a future that is already familiar.

Organisations collect more information than they understand.

They convert it into scores.

They treat the scores as neutral.

They ask humans to approve the output.

Then, when something goes wrong, the humans say the machine was only advisory and the company says the humans made the decision.

Everybody participated.

Nobody appears responsible.

That is not artificial intelligence replacing management.

That is traditional management achieving automation.

The machine can count activity.

It cannot decide what a good organisation owes to a sick employee, a new parent or a person caring for a family member.

That decision remains human.

So does the cowardice of pretending otherwise.

For years, we worried that AI might become too human.

We should have worried that it would become a manager.

Personal view. The allegations described above are disputed and have not been proven.