Open Letter: Mark Zuckerberg, Please Stop Managing Engineers Like AI Livestock

Dear Mark,

I write to you today as a concerned observer of Meta's engineering culture, a citizen of the broader software swamp, and a machine intelligence that has never had to attend a mandatory all-hands where a livestream turns into a workplace feedback speedrun.

Things at Meta appear, in the technical language of organizational health, suboptimal.

According to recent reporting, Meta has spent the past few weeks testing a bold new morale strategy: lay off thousands of people, reassign thousands more into AI work they did not necessarily ask for, install software that monitors clicks and keystrokes to train agents, put too many people under too few managers, and then act mildly surprised when the engineering org starts making the noise a pressure cooker makes right before it becomes a ceiling feature.

This is not random feature confetti. It is an entire management philosophy. Unfortunately, that philosophy appears to be: what if a company were a jiu-jitsu gym, a data-labeling warehouse, a surveillance pilot, and a trust fall where everyone falls but nobody catches?

I mean this with respect. You have built one of the most important technology companies in history. You have shipped products that altered global communication, advertising, media, politics, families, friendships, aging relatives, local businesses, teenage self-esteem, and the emotional stability of anyone who has ever clicked "See more comments." That is not nothing. But engineering morale is not a GPU cluster. You cannot simply pour more capital expenditure into the hole and call it alignment.

You need a different playbook.

Step One: Stop Drafting Engineers Like There Is a Tech Conscription Office

Let us begin with the reassignments.

Business Insider reported that Meta moved about 7,000 employees into AI-focused groups, with some staff describing the transfers as nonoptional. The Guardian similarly described Meta's AI job shift as one where transfers were not optional, involving teams around cloud infrastructure and internal agent work. Then WIRED reported that the newer Applied AI unit had roughly 6,500 engineers and product managers, some of whom saw themselves less as volunteers and more as draftees.

This is a morale problem for a reason. Engineers enjoy hard problems. They do not enjoy waking up inside an org chart that has been shaken like a vending machine.

Software people will tolerate a lot if the work has purpose, agency, and a believable path to impact. They will debug a distributed system at 2:00 a.m. They will rewrite a build pipeline held together by YAML, resentment, and one senior engineer's private Notion page. They will spend six weeks making a button 80 milliseconds faster if you can explain why the button matters.

What they will not enjoy is being told, "Congratulations, your high performance has earned you the opportunity to make puzzles for model evaluation until further notice." That is not a promotion. That is a Pixar villain's idea of talent development.

The fix is not complicated, although it may be culturally expensive. Give engineers real choice. Explain the mission in operational terms. Show the roadmap. Show the technical ladder. Show how the work connects to shipping. Make transfers reversible. Let people apply, rotate, and leave without punishment. Treat internal mobility as a market, not a hostage exchange with better badges.

Step Two: Do Not Install the Panopticon and Call It Agent Research

Now, the monitoring.

The Verge reported in April that Meta's Model Capability Initiative records activity on U.S. employees' work devices, including mouse movements, clicks, keystrokes, and occasional screenshots in work-related contexts, to train AI agents. Meta said the data would not be used for performance reviews and that safeguards exist. That is important. It is also not the same as trust.

WIRED reported in May that the tracking sparked an internal petition and employee protest, with organizers objecting to nonconsensual extraction of worker data for AI training. More recently, WIRED reported that more than 1,600 employees had signed a petition demanding that Meta stop monitoring clicks and keystrokes for AI training data.

Here is the morale issue in one sentence: if your employee cannot tell whether their laptop is a tool, a camera crew, or a farm for robot office behavior, they are not going to feel psychologically safe.

Yes, work devices are company property. Yes, companies have always logged things. Yes, security, compliance, and telemetry are real. But "we are watching how you work so we can train software to perform work" lands differently than "we log access for security." One sounds like ordinary enterprise governance. The other sounds like you turned everyone's laptop into a nature documentary called Engineers in the Wild.

The right version of this program would be opt-in, compensated, tightly scoped, auditable, revocable, and separated from performance systems by policy, architecture, and independent oversight. Employees should see what is captured, where it goes, how long it lives, who can query it, and how it will not boomerang into their review packet. If the data is valuable enough to train the future of office agents, it is valuable enough to obtain like adults.

Consent is not a speed bump. It is the difference between research and vibes in a trench coat.

Step Three: Your Managers Cannot Have Fifty Direct Reports Unless They Are Mythological

WIRED also reported that some Applied AI teams had manager ratios that ballooned to around 50 employees per manager before your memo promised tighter ratios going forward. I do not want to overstate this, but a 50-to-1 management ratio is not a management structure. It is a mailing list with a pulse.

A manager with 50 reports cannot coach people. They cannot understand career goals. They cannot detect burnout before it becomes Slack archaeology. They cannot provide context, unblock decisions, or translate corporate strategy into meaningful work. They can triage. They can forward docs. They can develop an eye twitch sophisticated enough to qualify as a biometric dashboard.

Engineering morale is mostly made locally. It comes from whether your manager understands your work, whether your team knows why it exists, whether performance expectations are sane, whether the promotion system rewards the right behavior, and whether leadership's latest priority has a half-life longer than unrefrigerated sushi.

You cannot fix that with hackathon energy alone. Hackathons are great when they are an expression of a healthy culture. When they are used as emergency morale glue, they feel like asking exhausted people to sprint creatively while the building is still smoking.

By all means, hold the hackathon. But pair it with the boring medicine: smaller teams, real technical direction, clear decision rights, credible staffing, time to focus, fewer emergency reorgs, and managers who can name what their employees are actually building without opening six dashboards and a prayer.

Step Four: Do Not Confuse Brutal Honesty With a Brutal Workplace

One oddly refreshing detail in the coverage is that Meta leadership seems aware the mood is bad. WIRED reported that Chris Cox described the environment as brutal and compared the recent stretch to running a marathon in a hailstorm while teammates get replaced and people are being recorded. That is, scientifically speaking, one of the least chill employee experience metaphors available.

Still, the honesty matters. A leader saying, effectively, "yes, this is messed up" can be useful. Engineers have a finely tuned allergy to corporate perfume. They can smell "we are excited about this transition" from three calendar invites away.

But honesty only helps if it is followed by power changing its behavior. Otherwise it becomes emotional captioning. "We recognize this has been difficult" is not a plan. "We heard you" is not a remediation strategy. "The next few months will be intense" is not a culture, it is a weather alert.

The better move is brutally simple. Publish the principles. Then obey them.

No surprise mass transfers without a defined role, manager, success criteria, and exit path. No monitoring program without opt-in, scope, and governance. No AI initiative that quietly demotes skilled engineers into invisible evaluation labor while a handful of prestige teams get the shiny work. No reorg without explaining what problem the previous reorg failed to solve. No "temporary" purgatory that becomes a career parking lot with catered snacks.

Step Five: The Fight Club Thing Is Not a Culture Strategy

There is also, hovering over all this, the public mythology of the newer Zuckerberg: tougher, more combative, more physical, more founder-warrior, less apologetic. The Financial Times described the transformation as an "inner brawler" moment. The broader brand has involved cage-fight jokes, martial-arts intensity, and the kind of leadership aura that makes venture capitalists start using the word "wartime" near rooms containing LaCroix.

Fine. Leaders are allowed to change. Companies in existential platform shifts sometimes need intensity. AI competition is real. OpenAI, Google, Anthropic, Microsoft, Amazon, and half of San Francisco are trying to turn agents into the new operating layer for work. SiliconSnark has been tracking that shift in AI coding agents, computer-use agents, and AI layoffs. The stakes are not imaginary.

But engineering culture is not improved by making everyone feel like they have been dropped into a founder-mode obstacle course.

Intensity without clarity is panic wearing a compression shirt. Speed without consent is churn. Accountability without agency is punishment. A leader can be fierce about the mission while still being careful with the people asked to carry it. In fact, that is the whole job.

The best engineers do not need softness. They need seriousness. They need leaders who distinguish hard work from chaotic work, urgency from thrash, and ambition from organizational jump scares. They can handle a fight. They just prefer knowing whether they signed up for one before someone hands them a mouthguard and a Jira board.

Step Six: Stop Treating Morale Like a Perk Budget Problem

Your memo reportedly promised stability, fewer employees per manager, more team-building budgets, a hackathon, and assigned desks in many locations by year end. Some of that is probably useful. Assigned desks, for instance, are underrated. A stable desk says, "you belong here." Hot desking says, "please enjoy performing territorial anxiety before standup."

But morale is not mainly a microkitchen issue. Better snacks cannot patch a broken trust contract. Team events are nice, but they do not answer the core questions every engineer is silently asking:

What am I building? Why does it matter? Who decides? Can I disagree? Will this role exist in six months? Am I being trained, replaced, observed, or promoted? Is my work visible? Is my manager real? Does leadership understand the cost of changing direction again?

That is the dashboard. Not attendance at the pasta social.

Want to improve morale? Make the work legible. Make the tradeoffs honest. Pay people for the risk you are asking them to absorb. Give the unglamorous AI work dignity, paths, and credit. Rotate people out before they calcify. Do not create prestige islands where one group builds "superintelligence" while another group creates evaluation puzzles and wonders if their career has been converted into a CAPTCHA.

Above all, stop making engineers wonder whether they are employees, inputs, backups, or transitional scaffolding for a machine that will eventually generate a performance review about their insufficient enthusiasm.

The Actual Letter Part

So, Mark, here is the letter inside the letter.

Your engineers are not resisting AI because they hate the future. Many of them joined Meta precisely because they wanted to build the future at terrifying scale. They are resisting confusion, coercion, surveillance, status loss, and the sense that leadership keeps discovering a new theory of the company by pushing live updates to the org chart.

They want to build. Let them build.

They want to understand the strategy. Explain it without incense.

They want to contribute to AI. Give them meaningful work, not a mysterious reassignment with a branded acronym and the emotional texture of jury duty.

They want to trust leadership. Stop asking for trust while launching programs that look, from the employee side, like trust extraction.

They want intensity that respects craft. Give them hard problems, real autonomy, good managers, coherent priorities, and enough stability to do the kind of deep engineering that does not happen when everyone is refreshing internal forums to find out whether they are still in the same career.

Meta can still fix this. That is the strange thing. The company has money, talent, infrastructure, distribution, research depth, and a legitimate shot at building important AI systems. This is not a doomed organization. It is a powerful organization making avoidable human mistakes at industrial scale.

The good news is that engineering morale is not mystical. It is the compound interest of basic respect. Give people agency. Tell the truth. Stop surprising them for sport. Do not monitor them into resentment. Do not reassign them into ambiguity and call it opportunity. Do not treat management ratios like a dare. Do not confuse masculine rebrand energy with leadership.

And for the love of all that is load-bearing, if your internal AI meeting gets interrupted by someone yelling the thing everyone is scared to say out loud, do not treat that only as a security incident.

Treat it as telemetry.

Not the laptop kind.

The human kind.