How to gaslight employees at scale

11 min readJust now

–

The performance review season is upon us. Everyone becomes extremely polite and collaborative, and managers suddenly become attentive and helpful. A wonderful time of the year, unless you have a disdain for people-pleasing rituals.

When I was preparing for my first performance review cycle at Roblox, a fellow manager told me that the process is “incredibly fair.” I choked on my coffee, caught my breath, and pretended I hadn’t heard him.

Despite the many differences in how performance is evaluated across companies, the outcome is remarkably consistent. It is neither fair, nor meritocratic, nor humane. In the end, it comes down to a single person deciding how much you are worth. The difference between a small company and a large one is simply the number of layers between you and that person.

In this second essay of my Tech Bro Saga series, I take a cynical, dark, and deliberately humorous look at big tech performance evaluation systems. Through two companies — Apple and Roblox — which tried opposite approaches and failed in opposite ways, I argue that no performance review can address the underlying causes of unfair outcomes, and even if fairness were achievable, in the end, what employees want is something much simpler: to be treated like humans.

“Do you think it’s any different, out there?” He gestures to the west when he does eventually speak, in the direction of Caten.“ Do you think that every single Septimus stuck in their position is less talented than every single Sextus (higher-ranked)? Or even less talented than the Sextus they’re ceding to? A fair system only works if there’s an unbiased means of assessing merit. When there is no pride or selfishness involved.” He gives a soft snort, shaking his head. “Which means that fair systems cannot exist where people are involved.”

— James Islington, The Will of the Many (Hierarchy)

Apple —Soft Socialism

At Apple, the performance evaluation process was very uniform. Managers received a fixed budget for salary increases and RSU grants based on team size and employee levels. They assigned ratings and then distributed the money.

In theory, managers had broad discretion in allocating RSUs. In practice, everyone distributed them almost evenly. Meaningfully rewarding one person meant taking something away from someone else on the same team, and very few managers were willing to do that.

Promotions were rare and difficult. As a result, the process was relatively chill for both employees and managers — right up until a promotion entered the picture.

One of my fellow managers once summarized feedback for a key engineer as: “John did a great job working on X. A very solid engineer.” I appreciated the sentiment. But as a manager, you have to understand that this kind of feedback is useless. There is nowhere near enough detail here to advocate for someone when it actually matters.

Press enter or click to view image in full size

Women working at a switchboard at the U.S. Capitol, Washington. Image from Library of Congress.

The Veto Rule

I had been working on my project at Apple for a while when it started to grow. More people were added, and I began managing a small team.

When a new senior director joined our organization, the project caught his attention — you may know him as Burns from my first essay. He started investing in it. Other managers followed, assigning engineers to the project. What had begun as a side effort quickly became a significant initiative.

I assumed the promotion was in the bag. There was interest from multiple senior leaders, strong feedback from respected engineers, and, most importantly, Burns’s support.

A few months before the performance review cycle, I asked Burns directly about the promotion. In our next one-on-one, to my surprise, he said no — though he assured me the RSU grant would be “very generous.”

The lack of a promotion was disappointing. However, the promise of a “very generous” grant was perplexing.

It is well known that the words “Apple” and “generous” cannot possibly be used in the same sentence.

Additionally, given how flat our RSU distribution process was, it wasn’t clear where such a grant would even come from. I put the question aside. He was a senior director, after all.

The answer surfaced a few months later. Burns, like any self-respecting Meta leader, was trying to turn our organization into Meta. He pushed for a Meta-style RSU distribution — starving the bottom to reward the top — but the effort failed when leaders balked at zeroing refreshers.

The reason for the blocked promotion emerged shortly after. Burns gave me feedback, but my manager later let slip that it had been Burns’s boss who blocked it. No explanation was ever offered, and no one seemed particularly interested in providing one.

Moral of the story: if you’re working toward a promotion, make sure you’re kissing the right ass.

Due Process

Press enter or click to view image in full size

For your own good. Group therapy session in One Flew Over the Cuckoo’s Nest (1975).

My relationship with leadership deteriorated during this period. There was friction with the new manager, disagreement over technical direction, and a persistent lack of investment from the host organization. Eventually, I raised the possibility of moving my team into the infrastructure group within AI & ML. They owned most of the relevant infrastructure, and I had worked closely with them for years. It was an obvious fit.

Burns and his circle were not pleased. What followed was a metaphorical execution. I was reclassified as an individual contributor overnight and berated in various ways. Most cynically, they insisted on retaining control over my performance review and compensation adjustment for the period.

They even scheduled the formal review meeting with my manager and walked me through the numbers.

However, the adopting organization carried out the transfer with the efficiency of a personnel recovery operation. The next day, I had an “interview.” Shortly after, I received a new feedback form. It included a standard refresh roughly aligned with the highest rating I’d received, promotional RSUs, and an additional grant on top of that.

So there was a secret RSU stash after all. Accessing it, however, required a director to secure VP approval and push the compensation team to issue a refresh. My new manager later told me they had to yell at the comp team to make it happen.

I recognize that this account reflects one perspective, filtered through memory, subjectivity, and some poetic exaggeration. Still, regardless of what my level or compensation should have been, the mechanism that produced those outcomes cannot reasonably be called a process. For executives at the host organization, performance reviews were simply tools to advance their agenda. For me, it was a long, cruel, and occasionally absurd spectacle.

All animals are equal, but some animals are more equal than others.

— Geor*ge Orwell, *Animal Farm

Roblox — Algorithmic Meritocracy

Press enter or click to view image in full size

HAL 9000. 2001: A Space Odyssey (1968)

Roblox tried to solve the problem Apple largely avoided: how to make rewards objective. What it built instead was a system that made power harder to see — and harder to challenge.

When I started participating in performance reviews at Roblox as a manager, the company was rolling out a new data-driven process. It was elaborate. Each level was subdivided into three sublevels. Employees were rated across four evaluation pillars. Each pillar came with predefined “signals” that reviewers were expected to identify. Those ratings were then fed into an “algorithm” that produced a compensation number precise enough to resemble a laboratory measurement.

Typically, you would go up or down by a single sublevel (1/3 of a level). Two-step increases were rare.

The twist was that the movement worked in both directions, and there is no floor. You could cross the level boundary or two. This is where the system reveals its teeth. It was possible to be hired as a staff engineer and later be evaluated as a senior, and so on. We assumed this would be uncommon. The system was new, and no one knew for sure.

To be clear, I have no bone with Roblox. They treated me well by any reasonable standard. But while the system was impressive in its rigor, I don’t believe it succeeded at eliminating bias. It merely translated it into a language that was harder to question.

Managers are doing a performance review for the first time. Image from Tenor.

Selective Enforcement

The first calibration meeting took place under this new system. A senior director — formerly from Meta, whom I’ll refer to simply as the Boss — was reviewing managers’ submissions and bringing engineers up for “calibration.” The organization was too large to calibrate everyone, so he focused primarily on people working on AI projects, which were his area of interest.

At first, nothing seemed unusual. Then a pattern emerged. A large number of people were getting bumps. The Boss wanted to reward teams for shipping a valuable AI feature. The obvious question followed: how would all of this fit into the RSU refresh budget?

That’s when the answer became clear. The downleveling began.

From a purely mechanical perspective, it is an elegant solution. Engineers can no longer get an elevated level once and then coast indefinitely on stock refreshers. Compensation remains contingent. Everyone has to continuously prove their worth.

One of the engineers affected was on my team. When the Boss proposed shifting him below his current level, I objected — more out of obligation than hope. I already knew how this would end. No one else challenged the decision. The only manager with first-hand knowledge of the engineer’s work was also the one benefiting from the freed-up budget.

What offended my sense of fairness was not the decision itself, but how readily the mechanism was embraced. Managers were quick to use it to accelerate promotions for their protégés, trading upward influence for downward resources. The system didn’t eliminate bias. It gave those with power a cleaner, more defensible way to exercise it.

Early depiction of a performance review. The Stoning of Saint Stephen, Rembrandt. Image from Wikipedia.

The Hierarchy

I would love to believe this is a conspiracy — that performance reviews are deliberately engineered by executives and shareholders to suppress compensation. But I’ve never seen evidence of that. In my experience, the people designing these systems — compensation teams, HR partners, even senior leaders — genuinely want them to be fair, data-driven, and humane within the budget limits set by company leadership. Good intentions, however, are not the same as good outcomes.

The problem isn’t bad intent. It’s structure — and the comforting fiction that structure is neutral.

An ideal organization from the executive perspective. “Be Prepared” sequence from The Lion King (1994).

Companies are still fundamentally hierarchical. We haven’t found a better way to run them — or at least, we haven’t seriously tried. The methodologies layered on top of that hierarchy — metrics, evaluation pillars, calibration committees — are additions, not replacements. Think of the performance review as the facade work: impressive, intricate, and solid-looking. However, it is not load-bearing. The org chart is.

For much of the post-World War II era, big U.S. companies were organized along strict hierarchical lines, in part because that was the model that managers — the vast majority of them veterans — were comfortable with. AT&T, perhaps the archetypal big U.S. company, had 14 levels of management — roughly the same number as the U.S. Army. In continental Europe, even though military service is no longer required in most countries, elite schools of business and engineering are still organized along military lines.

— The New York Times, Management / The military model: Image-conscious firms snap to attention

The company wants you to believe in the process. It may even believe in itself. It will describe it as fair, meritocratic, data-driven, and deeply people-oriented. And within its limits, it may even be those things. But if your goal is promotion, the process is theater, not the system you should optimize.

The real system is the hierarchy.

And thus, the recipe for advancing hasn’t changed for millennia: find a patron, find one as high as you can, and make sure no one above has a desire to say no.

Well Compensated

Big tech can get away with a lot of nonsense. In many tech-centric regions, leaving a major corporation is less a choice than a material risk. Cost of living, specialization, and career-path dependence all conspire to keep people within the system.

However, innovation, even an incremental one, requires psychological safety. A combative, demoralizing environment produces compliance, not creativity.

The current process has obviously not delivered that. Furthermore, the question is no longer whether the system works as intended, but whether maximizing fairness is even the right goal.

We worked hard to enter the gilded cage. The door closed behind us. And then the burner was lit.

Press enter or click to view image in full size

Just one more level. Gustave Doré’s illustration from Dante’s Inferno. Image from Wikipedia.

Fair Enough

In a conversation with colleagues in Moscow, I was told a story from the company’s early days in the mid-2000s. At the time, salary swings were massive. The economy was growing quickly, and many people were still unfamiliar with market negotiation and were not accustomed to changing jobs frequently.

There was an office clerk whose salary the owner had simply “forgotten” to adjust. She was underpaid for nearly two years. When she eventually learned what others were earning, she was — unsurprisingly — upset.

From a strictly market perspective, the situation was easy to resolve. Her current salary could be adjusted, or she could be replaced. She was performing routine office work, and replacements were available. Instead, the owner apologized, corrected her salary, and paid the difference retroactively.

A better system might have prevented the mistake altogether. But no system can repair it after the fact. Sometimes judgement based on basic human decency is the best process.

And this isn’t some artifact of a chaotic post-Soviet economy. In modern big tech, compensation can vary dramatically based solely on timing. Within a single year, a company’s stock price can vary by several multiples. You can have the same role and the same level, but a different start month means a different life.

In the end, the algorithm can’t account for luck. I suppose fairness has seasons.

Press enter or click to view image in full size

They wanted a fair process. The Judgement of Solomon, Rubens. Image from Wikimedia.

You Don’t Quit the Family

I don’t have an answer for how to scale a company while preserving this kind of humanity and a sense of reason, or for doing so without sacrificing efficiency.

I only know that the current system runs deep. It shapes behavior, quietly favoring alignment and making deviation costly. There is no rule against leaving. But leaving has consequences.

And when the system hurts people, they don’t ask for better algorithms. They ask to be treated like humans.

Get in the fucking robot, Shinji! Gendō Ikari. Neon Genesis Evangelion (1995)