Management by Vibes: A Field Guide

There is a sentence that gets delivered to senior engineers in large organizations, and it is delivered with the calm, regretful authority of a doctor reading test results that are fine, mostly, just career-shaped:

We can’t promote you because you haven’t worked on something that’s generated substantial business revenue.

I want to defend this sentence before I take it apart, because it deserves the defense. Unlike most of the language deployed at engineers during compensation season—the budget cycle, the finite number of headcount slots at each level, the calibration process that operates like weather—this one is not obviously nonsense. It’s not evasion. On the contrary, I’ve heard many women in tech hear from their managers that they need to “be more technical,” and I can express in words how bad-faith that feedback is and what an indictment it is of the person giving it. This, at least, has the shape of a criterion. It implies that somewhere, in some system, there is a number, and that the number could in principle be moved by me. I liked it. I want that on the record. After years of feedback with the epistemic status of a horoscope, here was something with a solvable structure, and I responded to it the way any engineer responds to a specification: I went and tried to build the thing. Which is where the trouble started, and the trouble is more interesting than the sentence.

The first step in demonstrating business impact is measuring business impact. Not arguing about it. Not asserting it in a promotion packet in the past tense, in prose, eighteen months after the fact, when everyone’s memory has quietly rewritten itself in their own favor. Measuring it, as it happens, in something close to real time. So the proposal was neither exotic nor expensive: embed real-time product analytics into everything we shipped on the platform. A first-party customer data platform. Instrumented flows. The full nervous system, so that we could see what people actually did with what we built rather than what we imagined they’d do—and, critically, see it early enough for the information to be actionable. The pitch had two halves, and I always thought the second half was the valuable one. First: find signal and build toward it—where usage is real, invest, deepen, double down, with evidence instead of enthusiasm. Second: find the absence of signal and stop. Kill the thing. Redeploy the humans onto something that has a pulse. I’d shipped some version of this at every company I’d worked at since making Staff—startup, mid-size, and enterprise. It’s not visionary. It’s plumbing. It’s the sort of thing that shows up in the first chapter of books executives put on their shelves.

I wrote the RFDs. I wrote the RAPIDs. I wrote enough DACIs that I have, at a level I’d describe as subconscious and mildly concerning, begun structuring non-work decisions as decision matrices. I took meetings with every director who’d sit still. This went on for years. And it went nowhere. I want to be exact about the failure mode, because “it was rejected” would be inaccurate and unfair. It was never rejected. Rejection is a decision, and a decision can be appealed, escalated, revisited with new evidence. This was something subtler and much harder to fight: it was agreed with, warmly, in every room, by everyone, and then not prioritized. Which is not the same as no. It’s the organizational equivalent of a man nodding along to your directions and then driving somewhere else entirely.

In retrospect I understand why, and the reason is worth stating plainly because it isn’t stupidity.

A measurement system is a machine that manufactures accountability, and it does not discriminate about whom it manufactures it for.

Instrumentation doesn’t just tell you which engineering teams are producing value. It tells you which bets were good. Which roadmaps were fantasy. Which strategy, championed by which person at which level, quietly failed to move anything at all. Nobody in the building would ever say out loud that they’d prefer not to know. Nobody had to. All that was required was for the ruler to remain, permanently, the third priority.

Meanwhile—and here is where this stops being a story about my career and starts being a story about the organization—initiative after initiative got greenlit with no measurement plan attached. No baseline. No target. No definition, agreed to in advance and written down, of what outcome would constitute success and what outcome would constitute a reason to stop. And when engineers asked the obvious question, the one they are literally trained and paid to ask—how will we know if this worked?—the question did not land as rigor. It landed as dissent.

I’ve thought about this conversion more than anything else in this essay, because I think it’s the master key. There is a specific cultural moment where “how will we measure this” stops being read as diligence and starts being read as insufficient enthusiasm—where the request for a ruler is received as an attack on the concept of buildings. And once an organization has made that conversion, a chain of consequences follows with something close to mechanical inevitability. First, the only acceptable response to a proposal becomes yes. Any position other than “let’s go” reads as hostility, so the informational content of agreement collapses to zero. Everyone says yes. Nobody has said anything. Second, since no bet has a defined success threshold, no bet can ever be formally declared dead. Initiatives don’t get killed; they get quietly de-staffed, which is the same outcome without any of the learning. Third—and this is the one that ought to keep people up at night—accountability doesn’t disappear. It migrates. It flows downhill and lands, in full, on the only people in the system whose output is concrete enough to be counted.

Which produces the arrangement I’d call the actual subject of this post. When an unmeasured initiative failed—as some fixed and entirely predictable percentage of unmeasured initiatives must, mathematically, gorgeously, forever—the engineering teams that built it were let go. The stated reason: they hadn’t delivered substantial business value. Read that against everything above. Success was never defined at the outset. The instrumentation that would have detected it was never funded. The engineers who asked for that instrumentation were coded as difficult. And then the absence of demonstrated value—an absence that was architecturally guaranteed by the organization’s own choices—was charged to their account, individually, at the moment of layoff. Endless liability. Zero upside.

That’s not a system. It’s a coin flip wearing a system’s coat, and the coat does not fit, and everyone in the room can see that the coat does not fit, and everyone in the room has independently concluded that the professional move is to compliment the coat.

I want to acknowledge the strongest counterargument here, because it’s real: some things genuinely resist measurement, and organizations that instrument everything can end up optimizing hard for whatever happens to be countable while the important, uncountable thing quietly dies. I’ve seen that failure mode. But it’s the failure mode of a mature measurement culture, and you cannot skip to the sophisticated critique of a discipline you haven’t practiced. You have to be able to count before you’re allowed to say counting isn’t everything. And note the trap the arrangement creates for anyone trying to advance: you can’t be promoted without demonstrating substantial business impact, and you can’t demonstrate business impact because the organization won’t measure anything. Both halves are defensible on their own. Together they form a closed loop with no exit, and the loop is nobody’s fault, which is exactly why it persists. Systems with no author are the hardest kind to fix.

For a long time I thought the cost of all this was promotions. It isn’t. Promotions are the least of it. Here’s what the absence of measurement actually does to a working environment, in ascending order of harm. Everything becomes equally urgent—the direct, mechanical consequence of having no data: without usage numbers, there is no way to know whether a bug affects nine people or ninety thousand, whether an opportunity is worth a quarter or an afternoon. Absent a ruler, every item defaults to maximum priority, because no one can prove otherwise and no one wants to be the person who guessed low. The result is a workforce operating permanently at the top of its stress range, not because the work is genuinely that critical, but because the organization has lost the ability to say that anything isn’t. Nobody develops—and this is the one I find hardest to write about without getting angry, because it damages people rather than roadmaps. Skill acquisition requires sustained context, staying with one hard thing long enough for it to reorganize how you think. An engineer in permanent firefighting mode does not accumulate depth. They accumulate surface: a thousand shallow contacts with a thousand systems, none of it consolidating into expertise. From the outside this looks like productivity. It is the exact opposite. It is an organization spending its people’s careers as fuel and recording it as velocity.

The maintenance surface becomes unpayable, because sprawl is what an org accretes when it starts many things and formally ends none of them. So the characteristic experience of a senior engineer in this environment is: being asked to patch a service nobody has touched in three years, which requires a critical dependency upgrade that ships breaking changes, while simultaneously on a drop-everything initiative that has been declared the most important thing in the world, by someone who is not aware of the first task and would not, if informed, change the declaration. And leadership does not acknowledge any of it. Not “fails to fix”—fixing is hard and I have sympathy for hard. I mean does not name it as a problem, in public, ever. And this is where I’d make my only genuinely harsh claim in this entire essay, a claim about revealed preference rather than about anyone’s character: an organization’s real values are not in its stated values. They are in the list of things it has decided not to fix. Every day that list stays the same, it’s a vote. And the vote wasn’t for customers, and it wasn’t for anyone’s career, and it wasn’t for anyone’s well-being.

I want to be honest that this doesn’t end with me changing the organization, because I didn’t. First, I redirected. If the system wasn’t going to compound my effort, I’d compound it myself. I started taking university coursework in machine learning—actual coursework, with actual problem sets, at an actual pace that made me feel elderly. I went through a mentorship program with one of our lead architects. And I started mentoring more junior and senior engineers myself, which turned out to be the single most reliable source of satisfaction available in the building, for the simple reason that its outcomes were legible to me immediately and did not require anyone’s approval to be real. Second, I stopped wanting the thing. Watching a full year of promotion cycles up close, the pattern was unmistakable: no quantitative targets, anywhere, for anyone.

Promotion was not a decision. It was a mood wearing a calibration meeting’s clothes.

And once I really absorbed that—not intellectually, but in the part of the stomach where dread actually lives—two things happened at once. The freeing one: I stopped feeling bad about being passed over, and stopped particularly coveting a judgment from a process I’d concluded was noise. The disappointing one, which took longer to arrive and hurt considerably more: an organization that operates this way can only ever win by luck. It cannot win by strategy, because it does not have one. A strategy is a set of falsifiable claims about the future.

Refuse to define the falsification conditions and you don’t have a strategy, you have a mood board with a budget.

Third, I stepped down as tech lead. Not as protest—protest assumes an audience, and I’d stopped believing in the audience. Simply: I’d been carrying a doubled workload in exchange for advancement that the system had now clearly told me was not a function of workload. So I put it down. The reasonable objection is that stepping down was capitulation—that I had standing and a platform and chose the quiet exit. I’ve turned that over a lot. My honest answer is that I’d spent years spending that standing on exactly this problem and had moved it approximately zero, and there’s a point at which continuing to spend a currency that buys nothing stops being principle and starts being a hobby. I’m not certain I’m right about where that point was.

And then something happened that I found so funny I’m still not sure I’ve fully metabolized it. I’d been in the role a long time. Long enough, I suspect, that people had gone numb to my voice in the specific way one goes numb to a housemate’s opinions—not disagreement, just familiarity, the ideas arriving pre-categorized as things we have already heard. New tech lead. Same team. Same platform. And a set of technical proposals that had been declined by our architecture group some months earlier went back in—same substance, same reasoning, different mouth—and were received as among the most compelling things anyone had heard all year.

Same words. Different mouth. Instant credibility.

I want to state clearly what I think this demonstrates and what it doesn’t. It doesn’t demonstrate that anyone was acting in bad faith; I don’t believe anyone was. What it demonstrates is that in the absence of an objective evaluation criterion, an organization will always fall back on a social one, because it has nothing else to fall back on. Novelty, standing, freshness, who said it, how tired the room is of that person—these aren’t corruptions of a rigorous process. They are what fills the vacuum where a rigorous process would otherwise be. Which is, in a roundabout way, the strongest argument for measurement I’ve ever encountered, and I got it for free, and I got it by leaving the room. I’m told the appropriate emotional response here is bitterness. I’ve found it, instead, clarifying, and a bit liberating: I now have fewer responsibilities, and when I do hold an opinion worth holding, I pass it to my successor and it gets heard as a new idea rather than as a rerun. I’m not a clout chaser, so I’ll take the arrangement.

Not long after all of this, the CEO and the CTO both resigned. I don’t know what it means. Nobody does yet; that’s what “resigned” means at the moment it happens. But I’ll offer this, and it’s the most hopeful sentence I’ve got: an organization can run on vibes for a remarkably long time. Longer than you’d think, longer than seems fair, right up until the moment the vibes leave the building along with the two people who were—deliberately or not—the ones supplying them. What replaces them will either define what winning looks like before the whistle, in numbers, in public, with a stated condition under which we’d admit we were wrong and stop—or it won’t, and we’ll go back to hoping.

Hope is a fine thing. It’s a terrible operating model.

I’d genuinely like to find out. I like this work. I love our customers, which is a sentence I’d like to be able to demonstrate with data rather than assert with adjectives. And I remain, against all available evidence and to my own continuing surprise, optimistic—a condition I’d describe as either resilience or a failure to update, and which I suspect will only be distinguishable in hindsight.