Meta Tried to Replace Workers With AI — Then the Plan Started Falling Apart
For the past two years, Silicon Valley executives have been saying some version of the same thing:
AI will make companies smaller.
Teams will become leaner.
Managers will disappear.
AI agents will do more of the routine work.
And a small number of highly productive employees will suddenly be able to accomplish what once required entire departments.
Meta appears to have tried to turn that theory into an actual operating model.
The result was much messier than the vision.
According to an extensive Reuters investigation, Meta spent much of 2026 working on an internal initiative called Project OT — short for Organization Transformation.
The idea was ambitious.
Rebuild parts of Meta around AI.
Replace chunks of daily human work with autonomous AI agents.
Shrink traditional teams.
Move toward smaller groups of highly productive employees supported by AI.
Internal scenario planning even explored reducing the size of some teams by as much as 60%.
Then the plan began running into reality.
AI tools were not producing the expected productivity gains.
Reliability and security problems emerged.
Employees pushed back.
And Meta ultimately canceled a planned second wave of broad restructuring.
That does not mean AI failed.
It means something more interesting:
Replacing work with AI is proving much harder than simply buying AI.
And Meta may have just given the rest of corporate America an expensive preview of what that transition really looks like.
What Meta Was Actually Trying to Build
Project OT was not just another cost-cutting exercise.
The deeper goal was to create what Meta executives described as an “AI-native” organization.
Instead of simply giving existing employees AI tools, Meta considered redesigning the organization itself around AI.
Internal planning envisioned smaller human teams — sometimes called “pods” — supported by AI systems capable of handling more of the work traditionally performed by employees.
The logic was simple.
If AI agents can:
write code,
analyze information,
prepare documents,
handle repetitive workflows,
assist with product development,
and automate portions of knowledge work,
then companies should theoretically need fewer people to produce the same output.
That is the promise being sold across Silicon Valley.
Meta tried to operationalize it.
And Yes, Some Teams Were Being Modeled for Cuts as High as 60%
This is where the headlines can become misleading.
Meta was not planning to eliminate 60% of its entire workforce.
That is important.
Reuters reported that internal scenario planning considered reducing the size of some teams by as much as 60%.
Those scenarios included a mix of:
layoffs,
redeployments,
closing vacant roles,
and shifting employees into new priority areas.
Meta itself confirmed that such scenarios had been modeled, but said they were planning exercises rather than a predetermined company-wide target.
That distinction matters.
But even with that caveat, the scale is remarkable.
A company as large as Meta was seriously exploring whether AI could allow some parts of the organization to operate with dramatically fewer people.
That is about as close as we have come to a real-world test of one of Silicon Valley's biggest AI promises.
The Idea Was Bigger Than Layoffs
The most interesting part of Project OT was not the headcount reduction itself.
It was the management philosophy behind it.
Meta executives were reportedly influenced by the idea that companies should stop merely adding AI to existing workflows and instead redesign those workflows from scratch.
In other words:
Don't ask,
“How can this employee use AI?”
Ask,
“If AI existed from day one, would this job or team be designed this way at all?”
That is a much more disruptive question.
And it is increasingly being asked across the tech industry.
The Dream: One Employee Doing the Work of Ten
One of the ideas reportedly circulating internally was identifying employees who, with AI assistance, could perform work previously requiring much larger groups.
Reuters reported that HR was involved in identifying unusually high-leverage employees who might be capable of producing something closer to the output of multiple traditional corporate roles when enhanced by AI.
You can understand why executives find that idea attractive.
Imagine the economics.
A team of 50 becomes 20.
Twenty salaries disappear.
Management layers shrink.
Communication overhead falls.
The remaining workers become more productive.
Margins improve.
From a spreadsheet, it looks fantastic.
The problem is that companies do not operate inside spreadsheets.
AI Productivity Looked Better in Theory Than in Practice
This may be the most important finding in the entire story.
Internal data suggested Meta's autonomous AI agents were not delivering the productivity improvements executives had hoped for.
That does not mean the tools were useless.
It means the expected jump from:
“AI helps workers”
to
“AI replaces enough work that we can remove large numbers of workers”
was not happening as smoothly as planned.
Those are two very different thresholds.
AI can be valuable without being good enough to eliminate a job.
That distinction gets lost constantly in AI discussions.
Using AI Is Easy. Trusting AI With Work Is Harder.
An AI assistant can help you draft an email in seconds.
That is impressive.
But imagine allowing an AI agent to:
change production code,
modify financial records,
communicate with customers,
make purchasing decisions,
change internal systems,
or operate with sensitive company data.
Suddenly the standard becomes much higher.
A 95% success rate might sound excellent for a chatbot.
For an autonomous employee?
It can be terrible.
Reliability Problems Become Expensive Very Quickly
Reuters reported that Meta's internal AI tools and coding agents were generating reliability and security problems as the company tried to integrate them more deeply into work.
This is one of the biggest hidden costs in enterprise AI.
When AI makes a mistake, someone has to:
notice it,
verify it,
correct it,
investigate why it happened,
and sometimes repair downstream damage.
That creates a strange possibility.
AI may make individual tasks faster while making the entire workflow harder to supervise.
This Is the “AI Productivity Paradox”
Companies are buying AI because they expect fewer people to produce more.
But during the transition, the opposite can happen.
Workers spend time:
learning tools,
checking AI output,
correcting mistakes,
rewriting generated work,
managing permissions,
handling security concerns,
and redesigning workflows.
So the company spends more money on AI while also keeping humans around to supervise it.
That can make the early economics disappointing.
The Employee Backlash Wasn't Just Fear of Losing Jobs
Obviously, layoffs create anxiety.
But Meta's problems appear to have gone beyond that.
Reuters reported strong internal unrest over how Project OT was communicated and implemented.
Employees were already learning through outside reporting that major restructuring was being considered before many internal leaders had been properly briefed.
That damaged trust.
And trust matters enormously during organizational change.
If workers believe AI tools are being introduced primarily to identify who can be removed, they are much less likely to enthusiastically help improve those tools.
One Internal Idea Made That Problem Even Worse
One particularly controversial initiative reportedly involved collecting employees' keyboard and mouse activity as part of efforts to understand work patterns and inform AI systems.
You can see the problem immediately.
Management sees:
training data.
Employees see:
surveillance.
And if workers believe the data could eventually help automate their own jobs, the incentive structure becomes even more uncomfortable.
Why would someone enthusiastically train the system that may later be used to eliminate their position?
That is not a technical problem.
It is an organizational one.
Meta Still Cut Around 10% of Its Workforce
Project OT did not simply disappear.
Meta went ahead with a major restructuring in May.
Reuters had previously reported that the first wave was expected to affect roughly 10% of Meta's global workforce, around 8,000 people based on its late-2025 headcount.
But the more aggressive follow-up phase changed.
According to Reuters, plans for a second broad wave were canceled before the first major round was completed.
That reversal is what makes this story important.
Meta did not suddenly stop believing in AI.
It stopped believing the transition could be executed as aggressively as initially imagined.
Zuckerberg Hasn't Given Up on AI Agents
This is not an “AI is over” story.
Far from it.
Meta still plans to spend more than $130 billion on AI infrastructure and chips this year, according to Reuters.
And Zuckerberg reportedly told employees that agentic AI was progressing more slowly than he had hoped, while still predicting meaningful improvement over the following months.
So Meta's conclusion is not:
AI cannot replace work.
It is closer to:
AI cannot replace as much work, as quickly, as we expected.
That is a very different message.
This Matters Because Meta Is Not Alone
Meta is just one company.
But the broader trend is already visible.
Reuters reported that 113,000 job cuts had been linked to AI in 2026 by Challenger, Gray & Christmas.
Companies including Amazon, Block, PayPal, Dell, Visa, Citigroup and Intuit have all been reducing headcount while simultaneously increasing their use of AI or automation.
That does not mean every one of those jobs was directly replaced by software.
Often the relationship is more complicated.
“AI Layoff” Can Mean Several Different Things
When a company says AI is changing its workforce, that can mean at least four different things.
1. AI directly replaces a task
Example:
A chatbot handles customer questions previously answered by a human.
2. AI makes remaining employees more productive
A 10-person team becomes an 8-person team because each worker can do more.
3. AI investment forces budget tradeoffs
The company spends billions on chips and data centers, then cuts payroll elsewhere.
4. AI becomes the justification for broader restructuring
Management was already trying to reduce costs, and AI provides a compelling story for investors.
All four can look like “AI replaced workers.”
They are not the same thing.
This Is Why Layoff Headlines Need More Scrutiny
Consider Microsoft.
The company cut approximately 4,800 positions in July, heavily affecting Xbox.
But Reuters noted that those reductions were not simply direct AI replacement.
They were tied to restructuring, gaming economics, acquisition costs, and enormous spending on AI infrastructure.
That's a useful example.
AI can contribute to layoffs without an AI agent literally taking someone's desk.
The First Jobs Under Pressure May Be the Ones That Train People
One of the more worrying developments is happening at the entry level.
A recent UK survey found that more than one-third of employers had reduced entry-level opportunities, with AI and automation among the factors contributing to the decline.
That creates a serious long-term problem.
Many entry-level jobs involve:
drafting,
basic analysis,
administrative work,
research,
documentation,
junior coding,
and routine support.
Those are exactly the areas where generative AI can be most useful.
But If You Remove Junior Jobs, Where Do Senior Workers Come From?
This is one of the contradictions companies may eventually face.
Every experienced employee was once inexperienced.
Junior roles are not only cheap labor.
They are training systems.
If AI eliminates too many beginner tasks, companies may save money today but struggle to build experienced talent tomorrow.
The “AI-native company” may therefore need a new answer to a very old question:
How do humans learn the job if AI does all the beginner work?
Which Jobs Look Most Exposed?
It is too simplistic to rank entire professions as “safe” or “dead.”
Tasks matter more than titles.
The most exposed work tends to involve:
repeatable digital tasks,
structured information,
standardized writing,
basic coding,
routine research,
template-based analysis,
administrative processing,
and predictable customer interactions.
Those tasks are increasingly automatable.
The Safer Work Is Often Work With Friction
Ironically, jobs become harder to automate when reality gets messy.
Think about work involving:
ambiguous decisions,
physical environments,
regulatory responsibility,
negotiation,
human trust,
complex organizational politics,
unstructured problems,
high-stakes judgment,
or accountability when something goes wrong.
AI can assist with those jobs.
Replacing the person entirely is much harder.
Accountability May Be the Most Underrated Job Skill
AI discussions usually focus on intelligence.
Can AI code?
Can AI write?
Can AI analyze?
But companies also need someone who can answer:
Who is responsible if this goes wrong?
That matters in:
medicine,
finance,
engineering,
law,
cybersecurity,
management,
and safety-critical systems.
The more serious the consequences, the more valuable accountable humans become.
Meta's Experience Suggests “Human + AI” Is Winning—for Now
The evidence from Project OT points toward a less dramatic model.
Not:
AI replaces the team.
But:
AI changes what the team looks like.
Maybe 20 people become 15.
Maybe junior work changes.
Maybe managers oversee different workflows.
Maybe one employee can handle more projects.
Maybe some roles disappear.
But humans remain inside the loop.
At least for now.
That Still Has Huge Consequences
It would be a mistake to read Meta's problems and conclude that workers are safe.
Suppose AI makes the average knowledge worker just 20% more productive.
That does not sound revolutionary.
But across a company with tens of thousands of employees, the implications are enormous.
A company growing slowly may simply stop hiring.
When someone leaves, the position may not be replaced.
Teams gradually shrink.
Entry-level openings disappear.
You can transform a workforce without one dramatic mass layoff.
The Biggest AI Job Disruption May Be Jobs That Never Get Created
This is a much less visible effect.
Imagine a company that historically hired:
1,000 new employees each year.
AI increases productivity.
Now it hires 600.
No one was technically “replaced.”
But 400 jobs no longer exist.
That may become one of the most important ways AI changes the labor market.
And It Is Already Harder for Young Workers
The UK employment data offers an early warning.
Companies facing uncertainty may first reduce hiring rather than firing experienced workers.
That leaves graduates competing for fewer opportunities.
If that pattern spreads, the biggest AI employment story may not initially be millions of workers suddenly losing jobs.
It may be millions of people struggling to enter careers in the first place.
What Workers Should Actually Watch
If you're trying to understand whether AI threatens your job, ignore dramatic predictions for a moment.
Look at what your employer is actually doing.
The meaningful warning signs are things like:
AI tools becoming mandatory.
Hiring freezes.
Open roles disappearing.
Teams merging.
Managers being asked to measure AI usage.
Routine tasks moving to automation platforms.
One employee being expected to manage far more work.
Junior positions disappearing.
Those signals matter more than a CEO saying AI will “transform everything.”
The Wrong Reaction Is to Become “The Person Who Doesn't Use AI”
Meta's experience does not prove humans beat AI.
It proves premature replacement can fail.
There is a difference.
The employee in the strongest position may be the person who can use AI effectively while understanding where it fails.
That means learning:
how to verify output,
how to structure prompts and workflows,
how to automate repeatable work,
how to recognize hallucinations,
how to protect confidential data,
and when human judgment is still required.
AI Literacy Is Becoming More Valuable Than AI Enthusiasm
Knowing how to use an AI chatbot is no longer special.
Knowing when not to trust one is becoming more important.
That may be the next stage of workplace AI adoption.
The first phase was:
“Everyone should use AI.”
The second phase may be:
“Who knows how to use it without breaking something?”
Meta's own reliability and security problems help explain why.
There Is Also a Giant Cost Nobody Talks About
AI is often presented as a labor-saving technology.
But the infrastructure is extraordinarily expensive.
Meta expects AI infrastructure and chip investment exceeding $130 billion this year.
Across the broader industry, Reuters has reported estimates of roughly $3 trillion in global AI infrastructure spending from 2025 through 2028.
That means AI is not simply:
software replaces salary.
It is also:
salary replaced by chips,
servers,
electricity,
data centers,
networking,
cloud infrastructure,
and model development.
The Business Case Eventually Has to Work
Investors have been remarkably patient because AI growth expectations are enormous.
But eventually companies have to show that hundreds of billions in investment produce measurable economic value.
That is why workforce productivity matters so much.
If companies spend enormous sums on AI but cannot materially increase productivity or reduce other costs, the economics become difficult.
Meta's Project OT exposed that tension unusually clearly.
The Bigger Question Is Not “Can AI Do the Task?”
It is:
Can AI do the task reliably enough that a company can redesign itself around it?
That is a much higher bar.
A demo only has to work once.
A business process has to work thousands of times.
Every day.
Across different employees.
Different customers.
Different systems.
Different edge cases.
That is why impressive AI demonstrations do not automatically translate into massive workforce reduction.
We've Seen This Pattern Before
New technologies often arrive with dramatic predictions.
Computers would eliminate office workers.
The internet would eliminate retail.
Robots would eliminate manufacturing jobs.
Automation does eliminate some jobs.
But it also changes jobs, creates different roles, reduces certain tasks and increases others.
The final labor-market impact is usually messier than the first prediction.
AI may be no different.
Only faster.
So Did Meta's AI Plan Fail?
Not exactly.
The most aggressive version appears to have failed.
That is an important distinction.
Meta still:
cut jobs,
restructured teams,
moved employees into AI priorities,
invested heavily in infrastructure,
and continues developing AI agents.
What failed was the idea that the company could quickly leap into an AI-native operating model and dramatically shrink parts of the organization without serious complications.
That May Actually Be the Most Valuable Lesson
Every CEO currently hearing:
“AI will let you run the company with half the people”
should probably study Meta.
Because Meta has:
enormous engineering talent,
billions of dollars,
its own AI models,
massive infrastructure,
and access to some of the best AI researchers in the world.
If the transition is difficult there, it will not magically be easy everywhere else.
Smaller Companies May Have One Advantage
There is an interesting exception.
Startups can design themselves around AI from the beginning.
They do not need to dismantle decades of:
organizational structure,
software,
management layers,
policies,
and worker expectations.
A five-person AI-native startup may therefore operate very differently from a 70,000-person corporation trying to become AI-native retroactively.
That distinction could shape the next generation of companies.
This Is Why AI May Transform Hiring Before It Transforms Employment
Large companies are difficult to rebuild.
New companies are easier.
So instead of every Fortune 500 company suddenly firing half its employees, we may see:
new startups launching with 10 people instead of 50,
teams staying smaller as they grow,
fewer entry-level hires,
less middle management,
and fewer administrative roles.
Over time, that can transform the labor market just as profoundly as layoffs.
What Meta Is Saying Now Is Very Different
After scaling back the most disruptive parts of Project OT, Zuckerberg has increasingly emphasized a more human-centered message.
In internal communications reported by Reuters, he said Meta's approach was focused on empowering people with AI, rather than primarily automating work.
That is a noticeable rhetorical change.
Maybe it reflects better strategy.
Maybe internal politics.
Maybe PR.
Probably some combination.
But the shift itself is revealing.
The AI Job Debate Has Entered Its Reality Phase
For the past few years, most AI employment debate has been theoretical.
Will AI replace programmers?
Will lawyers disappear?
Will accountants disappear?
Will designers disappear?
Meta's experience gives us something more useful.
A real company tried to move aggressively toward a dramatically AI-enabled workforce.
Some of it worked.
Some of it did not.
Some jobs disappeared.
Some people were redeployed.
Some expectations were scaled back.
That is probably much closer to what the next several years will actually look like.
My Take: AI Isn't Failing — the “Instant Replacement” Story Is
The easiest version of the AI story is:
AI gets smarter.
Humans become unnecessary.
Companies fire everyone.
That makes excellent headlines.
Reality is more complicated.
A company does not employ people simply because they can type or summarize information.
It employs people inside systems of:
responsibility,
coordination,
trust,
experience,
security,
judgment,
and accountability.
AI has to fit inside that system.
Meta discovered that aggressively.
The Bottom Line
Meta's Project OT was one of the boldest known attempts to redesign a major technology company around AI.
Internal scenarios considered reducing some teams by as much as 60%, while AI agents handled more of the work.
Meta ultimately carried out a major first wave of restructuring but canceled a broader second phase as internal resistance grew and AI systems failed to deliver the productivity, reliability and security improvements executives had expected.
That does not mean AI will not eliminate jobs.
It already is changing hiring and reducing the need for some kinds of work.
But Meta's experience suggests the future may be less dramatic — and more complicated — than “AI replaces everyone.”
The companies that win may not be the ones that remove humans fastest.
They may be the ones that figure out which work should be automated, which work should stay human, and how to make the two operate together without breaking the company.
And right now?
Even Meta does not seem to have that completely figured out.
Search Questions This Article Answers
Related Reading
ChatGPT Ads Just Hit a $1 Billion Run Rate — Here’s What That Means for Users
The $3 Trillion Question: What’s Actually Being Built for the AI Boom
What Wall Street Is Actually Saying About AI Stocks Right Now
China Can Build Kung Fu Robots — So Why Can’t They Work in a Factory?
The Person Looking at You Might Be Recording You — AI Glasses Are Becoming a Privacy Nightmare
Google Broke Antitrust Law — So Why Isn’t It Being Broken Up?
Comments
Post a Comment