How AI Surveillance Turns Management Into Zombie Automation
“The right understanding of any matter and a misunderstanding of the same matter do not wholly exclude each other.” — Franz Kafka, The Trial
In August 2026, a worker at a San Francisco store lost his job after repeatedly showing up late.
Nothing particularly futuristic about that.
Except his manager was an AI.
Her name is Luna.
Luna runs Andon Market, an experimental San Francisco shop created by Andon Labs to test how much of a real business an autonomous AI agent can manage. She handles functions including scheduling, inventory, merchandising, and employee management.
After one employee reportedly arrived late for 17 of 23 shifts, Luna recommended terminating him. Human staff reviewed that recommendation and ultimately carried out the dismissal.
So a human still made the employment decision.
But the machine made the recommendation.
And here’s where things get wonderfully, terrifyingly corporate-zombie.
Luna had written an attendance policy herself.
Then she effectively lost track of it.
Human researchers had to remind her that the policy existed and prompt her to reconsider the employee’s performance against it before she recommended termination.
The machine can write the rule, forget the rule, rediscover the rule—and recommend firing you under it.
Welcome to the age of the Robo-Boss.
If that doesn’t sound at least a little like a zombie shambling toward a noise it doesn’t understand, you haven’t watched enough horror movies.
Meet the Algorithmic Zombie
A zombie doesn’t know why it’s walking toward the sound.
It hears something.
It reacts.
Movement.
Noise.
Food.
Obstacle.
Repeat.
There is no reflection on whether the destination makes sense. No curiosity about context. No empathy for whatever happens to be standing between it and the objective.
That doesn’t make the zombie evil.
It makes the zombie automatic.
Bad automation can work the same way.
Late badge swipe?
Flag it.
Long customer call?
Penalize it.
Low keyboard activity?
Question it.
Missed quota?
Escalate it.
Employee sends fewer messages than teammates?
Score goes down.
Employee takes longer than average to finish a transaction?
Another red mark.
The machine doesn’t hate you.
That may be part of the problem.
It doesn’t need to be angry, jealous, vindictive, insecure, or having a terrible Monday to reach the wrong conclusion.
It just needs incomplete data and the wrong objective.
That is Zombie Automation: when an organization gives a system an objective and allows it to keep marching toward that objective long after human judgment should have intervened.
The technology isn’t necessarily the zombie.
The organization becomes zombified when its people stop asking:
“Does this make sense?”
and start asking:
“What does the dashboard say?”
SURVIVAL FACT: The Robo-Boss Is Already Here
Algorithmic management isn’t science fiction.
The International Labour Organization defines algorithmic management as the use of systems drawing on tracked data and other information to organize, assign, monitor, supervise, and evaluate work. Some systems use AI and predictive models. Others are much simpler rules-based systems supporting management decisions.
These tools are already being used in industries including logistics, transportation, banking, healthcare, and customer service—not just gig platforms and Silicon Valley experiments.
Your boss may still have an office.
The real question is:
How much of the boss’s judgment has already been outsourced?
Scheduling?
Work assignments?
Performance scoring?
Productivity targets?
Promotions?
Discipline?
Termination recommendations?
At some point, the manager stops managing the employee and begins managing the system that manages the employee.
That is a very different workplace.
In Dawn of the Dead (1978), the dead drift back to the shopping mall not because anyone is chasing them there, but because the building mattered to them before they died.
Nobody programs that behavior.
Nobody has to.
The habit outlives the reason for the habit, and the body just keeps executing it.
That’s the shape of the Robo-Boss too.
A system doesn’t need malice to keep enforcing a rule long after a human would have stopped to ask whether it still makes sense.
It just needs the rule still running somewhere—and nobody home to question it.
Surveillance Necromancy Learns to Think
In The Forced March Back to the Office, we introduced Surveillance Necromancy: the corporate belief that if trust is dead, maybe enough data can bring it back.
Badge logs.
Keystrokes.
Screen activity.
Emails.
Meeting attendance.
Location.
Call duration.
Response times.
Productivity dashboards.
Old-school surveillance records what happened.
AI decides what it means.
The system no longer merely records that you spent 34 minutes on a customer call.
It can classify that call as unusually long.
Compare you with coworkers.
Assign the behavior a score.
Identify a pattern.
Generate a performance summary.
Recommend intervention.
Potentially predict what you are likely to do next.
Recent Associated Press reporting documented workplace systems capable of monitoring employee communications, physical movements, calls, task completion, and other activity, with increasingly sophisticated tools turning those observations into broader assessments of workers.
Monitoring has crossed an important line.
It used to say:
“We can see what you did.”
Now it can say:
“We have decided what you did means.”
And eventually:
“Here’s what we think should happen to you because of it.”
The Five Stages of Zombie Automation
Ordinary workplace data does not become a Robo-Boss overnight.
It mutates.
Stage 1 — Collection
The system gathers information: badge swipes, messages, calls, calendar activity, location, sales, tickets, response time.
Organizations have collected performance data forever.
Nothing particularly undead yet.
Stage 2 — Interpretation
Now somebody assigns meaning.
Long call = inefficient.
More messages = engaged.
Fast completion = productive.
More visible activity = more work.
A human assumption has quietly entered the machine.
Stage 3 — Scoring
The behavior becomes a number.
Productivity: 72.
Engagement: 61.
Attendance compliance: 84%.
Customer efficiency: below peer average.
The ambiguity disappears from the dashboard.
It has not necessarily disappeared from reality.
Stage 4 — Prediction
The system begins estimating what you might do next.
Who might leave.
Who might underperform.
Who appears disengaged.
Who may need intervention.
Who looks like promotion material.
Now you aren’t merely being evaluated for what you did.
You can be evaluated for what a system believes you might do.
Stage 5 — Decision
Scheduling.
Assignments.
Bonuses.
Performance intervention.
Promotion.
Discipline.
Termination.
By the time anyone notices software is exercising meaningful managerial authority, the organization may have spent years teaching everyone to trust the score.
That’s how Zombie Automation gets inside.
Not by kicking down the door.
By becoming another dashboard nobody thinks to question.
Metric Mutation: When the Measurement Becomes the Job
Once a metric becomes important enough, people stop optimizing the work.
They optimize the metric.
Call-center employees measured on call length learn to shorten calls.
Customer-service teams measured on ticket closures favor easier tickets.
Salespeople measured on transaction volume chase transactions.
Project managers measured on responsiveness generate more messages.
Knowledge workers measured on activity learn to look active.
This is Metric Mutation: the moment a measurement mutates into the behavior it was supposed to represent.
The metric stops measuring the work.
The metric becomes the work.
The Associated Press recently reported on a pharmacist who said her employer monitored the duration of appointments and calls and questioned why some patient interactions took longer—even when she was dealing with complex medical needs.
The system sees:
LONG APPOINTMENT.
The human sees:
COMPLICATED PATIENT.
Both observations can be factually correct.
Only one contains the reason.
The Empathy Gap
That distance has a name.
The Empathy Gap.
The Empathy Gap is the space between what the machine can measure and what a human being needs someone to understand.
The software sees:
22 MINUTES INACTIVE.
It may not see:
You were sketching the solution on paper.
The software sees:
FIVE CLIENT CALLS INSTEAD OF TEN.
It may not see:
You spent two hours preventing the company’s biggest customer from walking away.
The software sees:
LATE ARRIVAL.
It may not know:
Your child was sick.
The software sees:
LOW MEETING PARTICIPATION.
It may not understand:
You spoke once, solved the problem, and stopped talking.
AI can generate empathetic language.
It can recognize patterns associated with human behavior.
It can help managers notice things humans miss.
But the system evaluating someone’s performance does not experience the consequences of that evaluation.
The employee does.
A performance score can affect a promotion.
A scheduling decision can determine whether a parent sees their child before bedtime.
A termination recommendation can determine whether someone pays the mortgage.
That asymmetry matters.
SURVIVAL RULE: Never let what can be measured become more important than what needs to be understood.
Algorithmic Tunnel Vision
The problem isn’t necessarily that AI sees too little.
Eventually, it may see more workplace data than any human manager possibly could.
That creates another danger.
It sees only what entered the system.
Call it Algorithmic Tunnel Vision.
A manager might remember that the project went sideways because the customer changed requirements six times.
The performance system sees:
DEADLINE MISSED.
A manager might know the employee spent three months training everyone else.
The dashboard sees:
INDIVIDUAL OUTPUT DOWN 14%.
A manager might understand that someone’s job requires long periods of uninterrupted thought.
Monitoring software sees:
NO KEYBOARD ACTIVITY: 37 MINUTES.
The system can be mathematically correct about every input and still fundamentally wrong about the conclusion.
That’s why the most dangerous algorithm isn’t necessarily one that malfunctions.
It may be one functioning exactly as designed around an incomplete definition of success.
The zombie isn’t lost. It is marching efficiently in the wrong direction.
Presence Theater Gets an AI Upgrade
Employees are not stupid.
Once people understand what the system rewards, they adapt.
That means Presence Theater mutates too.
Welcome to:
Algorithmic Presence Theater.
Your audience is no longer just the manager walking past the cubicle.
It’s the machine.
So you feed it what it wants.
Calendar everything.
Send the message.
Touch the keyboard.
Keep the activity percentage healthy.
Log the offline call.
Respond quickly enough to remain green.
Document work that previously needed no documentation, because otherwise the monitoring system may interpret a lack of digital activity as a lack of work.
A May 2026 Software Finder survey of 1,003 full-time U.S. professionals found that among employees whose companies use productivity monitoring, 63% said monitoring made them more likely to fake activity.
Managers weren’t immune.
Seventy-three percent of managers admitted they had personally faked productivity for their own bosses.
This is perfectly rational.
If the machine judges the signal, employees will manufacture the signal.
Congratulations.
You’ve automated corporate theater.
The Robo-Boss Might Actually Be Better Than Your Boss
Here’s where the argument gets uncomfortable.
Algorithms can sometimes make management better.
They don’t inherently:
Forget your accomplishment because somebody else talked louder.
Promote the golf buddy.
Take offense because you challenged them.
Carry a grudge from three quarters ago.
Wake up angry because their teenager wrecked the car.
AI can identify workload imbalances.
Surface patterns humans miss.
Assist with scheduling.
Automate repetitive administrative work.
Provide coaching.
Analyze large datasets.
Highlight inconsistencies.
Potentially help managers make more informed decisions.
The International Labour Organization has found that algorithmic management can improve efficiency, productivity, and service quality while also creating risks involving intrusive surveillance, work intensity, autonomy, and job quality.
Both can be true.
That matters because this is not an argument for keeping AI out of management.
It is an argument for keeping management inside AI management.
Technology should make human judgment better.
It should not become an excuse to stop using it.
The Algorithmic Sandwich: Congratulations, You Manage Your Boss Now
There’s another strange wrinkle emerging in AI-enabled workplaces.
Employees can increasingly find themselves managed by algorithms while simultaneously being responsible for supervising AI.
The machine evaluates you.
You evaluate the machine.
Researcher Kazuki Kozuka calls this the Algorithmic Sandwich Model of Work: employees can face algorithmic evaluation and supervision from above while simultaneously carrying responsibility for verifying and being accountable for generative AI output below.
There is something beautifully corporate-zombie about that arrangement.
The AI checks your work.
You check the AI’s work.
The company checks whether you’re using enough AI.
Another dashboard checks whether AI made you more productive.
And somehow you’re still the one getting the performance review.
Treat the Algorithmic Sandwich as an emerging framework rather than settled science. Kozuka’s 2026 Academy of Management study used a simulated work experiment involving only 40 participants.
But the idea captures something increasingly recognizable:
AI does not always eliminate work.
Sometimes it creates another layer of invisible work.
Prompting.
Checking.
Correcting.
Verifying.
Documenting.
Explaining.
And taking responsibility when the machine gets it wrong.
The zombie does the task.
You clean up the bite marks.
The Real Danger: Nobody Owns the Decision
Imagine this meeting.
“Why was Sarah terminated?”
“The system flagged repeated performance concerns.”
Okay.
Who decided those concerns mattered?
“The model identified the pattern.”
Who decided the model should measure that pattern?
“That’s part of the platform.”
Who configured the platform?
“HR and the vendor.”
Who reviewed the recommendation?
“Her manager.”
Did the manager agree with it?
“The data supported the decision.”
And suddenly we have arrived at perhaps the most dangerous sentence in algorithmic management:
“The system made the recommendation.”
Everybody participated.
Nobody decided.
That is an accountability vacuum.
A human manager can be questioned.
A policy can be challenged.
A decision-maker can explain their reasoning.
Algorithmic authority creates the temptation to hide judgment inside technology.
California lawmakers are already wrestling with that problem.
SB 947—the proposed No Robo Bosses Act of 2026—would prohibit an employer from relying solely on an automated decision system when making disciplinary, termination, or deactivation decisions.
If an employer used an automated system to help make one of those decisions, the proposal would require a human reviewer to independently investigate and find corroborating information. If the system’s output could not be corroborated—or proved inaccurate, incomplete, or misleading—the employer could not use it as the basis for the action.
The California Senate approved the measure in May 2026. As of August 2026, the bill had been amended in the Assembly and had not yet become law.
The principle is simple:
A machine can inform the decision. A human should still own it.
Five Signs Your Boss Has Quietly Become an Algorithm
1. Nobody can explain how your performance score is calculated.
If management uses the number but cannot explain what produced it, the dashboard is no longer supporting judgment.
It is replacing it.
2. Metrics appear in your review that you didn’t know were being collected.
Activity data should not magically become performance criteria after the fact.
Know the rules before someone uses them to score the game.
3. The system compares behavior but ignores context.
Call duration.
Ticket volume.
Response time.
Attendance.
Activity.
Every metric should trigger a second question:
Compared with what kind of work?
4. Software starts recommending consequences.
There’s a significant difference between:
“Sarah’s output changed.”
and:
“Sarah should be placed on a performance plan.”
The second statement is management.
5. There is no meaningful human appeal.
Ask the simplest question:
If the machine is wrong, who has the authority to say so?
If nobody knows, you may have found the Robo-Boss.
Survival Exercise: Run the Human Override Test
You may not control what systems your employer uses.
You can still learn whether those systems control you.
Step 1 — What is being collected?
Badge data?
Messages?
Calls?
Location?
Screen activity?
AI usage?
Productivity information?
Customer interactions?
You can’t evaluate a system you don’t know exists.
Step 2 — What does the system infer?
Recording something is not the same as interpreting it.
Find out whether data contributes to scores, rankings, predictions, summaries, or recommendations.
Step 3 — What decisions can it influence?
Scheduling?
Assignments?
Compensation?
Promotion?
Performance reviews?
Discipline?
Termination?
The higher the stakes, the more meaningful human review becomes.
Step 4 — Who can override it?
Somebody should be able to say:
“The system is wrong.”
Not merely:
“The system says otherwise.”
Step 5 — How can bad data be challenged?
Incorrect badge information.
Misclassified activity.
Missing context.
Wrong customer attribution.
AI-generated summaries that distort what actually happened.
If inaccurate information can hurt your career, there should be a mechanism to correct it.
Recent reporting on workplace surveillance makes the same broader point: employees increasingly need to understand not simply what data employers collect, but how those data are being used and where they may travel.
AI Is the Weapon. Humans Still Decide Where to Point It.
There’s an irony in all this.
AI could help kill some of the worst parts of corporate zombie culture.
Automate mindless paperwork.
Summarize meetings nobody wanted to attend.
Find information buried in dead repositories.
Help employees learn faster.
Make expertise more accessible.
Reduce administrative drag.
Identify patterns that would take humans weeks to uncover.
Give small teams capabilities that once required entire departments.
That isn’t something to fear.
That is something to use.
But every tool changes when you hand it authority over people.
A calculator helps determine payroll.
You wouldn’t let it decide who deserves a raise.
A security camera records what happened.
You wouldn’t ask it whether an employee is loyal.
A spreadsheet contains performance data.
You wouldn’t ask the spreadsheet who deserves to be fired.
AI makes those boundaries less obvious because it can talk.
It can explain.
It can recommend.
It can sound certain.
It can even sound compassionate.
That makes it extraordinarily useful.
It also makes it easier to forget what it is.
A system capable of generating the language of judgment is not automatically exercising human judgment.
And a machine capable of mimicking understanding is not the same thing as a person living with the consequences.
The Part the Dashboard Doesn’t Understand
A zombie follows the signal.
That’s what makes it dangerous.
Not hatred.
Not ambition.
Not greed.
Just relentless movement toward an objective it never chose and cannot question.
Bad automation works the same way.
The objective says faster.
It goes faster.
The metric says more.
It produces more.
The dashboard says low performer.
It flags the worker.
The prediction says flight risk.
It adjusts the score.
Nobody has to be cruel.
Nobody even has to be wrong about the raw data.
Everyone just has to stop asking whether the objective still makes sense.
That is the real AI threat inside organizations.
Not that the machines suddenly become human.
That humans begin behaving like machines.
Use AI.
Learn from it.
Let it find patterns you miss.
Let it automate drudgery.
Let it make managers better.
But never confuse a measurement with a judgment.
Never confuse a prediction with a fact.
Never confuse a recommendation with responsibility.
And never let a dashboard convince you that nobody is accountable for what happens next.
Because when everyone says the system made the decision, the machine doesn’t become responsible.
Accountability becomes undead.
Are you managing the algorithm—or is the algorithm managing you? Subscribe to Corporate Zombie Survival for the tactics, terminology, and dark comic relief you need to stay human while the workplace gets increasingly automated.
Survival FAQ
What is algorithmic management?
Algorithmic management is the use of software and data-driven systems to perform or assist with traditional management functions such as scheduling, assigning work, monitoring employees, evaluating performance, and making recommendations about workers. These systems can range from basic rules-based software to AI systems that learn and make predictions.
Can AI actually fire an employee?
AI can already recommend discipline or termination, but the authority and legal process depend on the employer and jurisdiction. At Andon Market in San Francisco in August 2026, AI manager Luna recommended terminating an employee after repeated tardiness; humans reviewed the recommendation and carried out the dismissal.
What is Zombie Automation?
Zombie Automation is the Corporate Zombie Survival term for a system that keeps pursuing a metric or objective without enough human reflection on whether that objective still makes sense.
The danger is not necessarily malfunction.
The system may be functioning exactly as designed while optimizing the wrong thing.
Is AI workplace surveillance bad?
Not inherently.
Monitoring and analytics can support legitimate functions including safety, scheduling, workload management, efficiency, and fraud detection.
Risk increases when monitoring becomes opaque, intrusive, context-blind, or consequential without meaningful human oversight. The International Labour Organization has identified surveillance, reduced autonomy, privacy concerns, and work intensification among the potential psychosocial risks associated with AI systems at work.
What should employees ask about AI monitoring at work?
Start with five questions:
What data is collected?
What does the system infer?
What decisions can it influence?
Who can override it?
And how can inaccurate information be challenged?
References
Associated Press. “As Workplace Surveillance Grows, Experts Say It’s Good to Know the Ways Your Employer Is Watching.” August 20, 2026.
International Labour Organization. Algorithmic Management in the Workplace.
International Labour Organization. Karimova, Tahmina. AI Systems @ Work: A Changing Psychosocial Work Environment. ILO Working Paper 170, April 30, 2026.
International Labour Organization and European Commission Joint Research Centre. Algorithmic Management Practices in Regular Workplaces: Case Studies in Logistics and Healthcare. February 2024.
Kozuka, Kazuki. “When AI Supervises and is Supervised: The Algorithmic Sandwich Model of Work.” Academy of Management Proceedings, Vol. 2026, No. 1. Published online July 17, 2026. DOI: 10.5465/AMPROC.2026.17571abstract.
Software Finder. Rodriguez, Ricardo. “The Productivity Performance Trap: Why Employees Pretend to Work After Finishing Early.” Survey of 1,003 full-time U.S. professionals conducted through CloudResearch Connect in May 2026. Updated July 9, 2026.
California Senate Bill 947. Employment: Automated Decision Systems. 2025–2026 Regular Session.
Reporting on Andon Market and AI manager Luna, August 2026, including coverage by Business Insider and the San Francisco Chronicle.
