What Would You Say You Do Here?
My best engineers cannot explain what they do. Same goes for my best account managers, and for the ops lead who somehow always knows when a team is about to implode.
Amazon just made it a formal performance review requirement: list 3-5 specific accomplishments, with examples of what you delivered and how you'll keep leveling up.
A year ago, DOGE tried the same thing with the entire federal workforce. It didn't end well.
In February 2025, the Office of Personnel Management sent every federal employee an email with instructions straight out of a Mike Judge film. List five things you accomplished last week, reply by Monday, and failure to respond would be treated as a resignation. Within hours, the FBI, DOJ, State Department, Pentagon, and half a dozen other agencies told their staff to ignore it. The program quietly died out over the following months, but the thinking behind it never went away.
In 1999, Mike Judge wrote a scene that has haunted corporate America ever since. Two efficiency consultants named Bob sit across from a middle manager and ask: "What would you say you do here?"
Tom Smykowski stumbles. He gets defensive. He eventually screams that he has people skills, as if volume could substitute for substance. The conventional reading is that Tom is a useless bureaucrat who deserves what's coming.
There is another reading. Tom isn't useless but inarticulate, and those are not the same thing.
In 1966, the philosopher Michael Polanyi opened his book The Tacit Dimension with a line that should be carved into the walls of every HR department: "We can know more than we can tell."
Polanyi wasn't defending bureaucratic bloat. He was describing a fundamental feature of human cognition. We recognize faces without being able to explain how. We ride bicycles without being able to derive the physics. We develop expertise through thousands of hours of practice and then lose the ability to explain what we learned.
MIT economist David Autor formalized this in 2014 as "Polanyi's Paradox," arguing that tasks requiring flexibility, judgment, and common sense remain the hardest to automate precisely because we understand them only tacitly. The knowledge exists, but it resists articulation.
The senior engineer who "just knows" which edge cases will cause production outages. The account manager who can read a client meeting and sense when a deal is souring. The operations lead who spots early signs of team dysfunction before anyone else does.
None of them can write a job description that captures what they actually do. Their value exists in the space between formal procedures and actual outcomes. Ask them to describe their contribution in five bullets and you'll get something thin and generic, or you'll get nothing at all.
Both the DOGE email and Amazon's new requirement reveal something about how their architects think about work. They presuppose that valuable work is describable work, that articulation equals contribution, that inability to explain is evidence of inability to perform.
This rewards a specific kind of employee: the one who is good at self-promotion. The worker who filed impressive-sounding status updates looks productive on paper. The worker who prevented three crises last month by catching problems before they escalated has nothing flashy to report, no bullets or deliverables, just disasters that didn't happen.
There's a class dimension here that deserves attention. Education systems train some people to articulate their value and other people to provide it. The ability to write persuasively about your accomplishments correlates with certain kinds of schooling, certain kinds of jobs, certain kinds of practice. A GS-7 processing benefits claims may have developed judgment that runs deep about which cases need escalation and which can be resolved quickly. The knowledge is real. The vocabulary to describe it may not be.
This isn't an efficiency measure but a literacy test for bureaucratic performance.
The timing makes it worse, and it's not subtle. By 2030, all 76 million baby boomers will have reached retirement age. Ten thousand are turning 65 every day. According to APQC research, organizations expect 51% of their workforce to retire or leave within five years. Most of what they carry with them will never be captured.
Experts estimate roughly 80% of organizational knowledge is tacit and undocumented. According to research from IDC, Fortune 500 companies lose roughly $31.5 billion annually due to ineffective knowledge sharing. As the boomer exodus accelerates, that cost can only grow.
The people who can't explain what they do are often the people who hold the organization together. The institutional memory, the contextual judgment, the sense of how things actually work rather than how the org chart says they should. All of that walks out the door when they do.
DOGE tried the Office Space playbook and quietly abandoned it. Now Amazon is running the same play. The question isn't whether the demand sounds reasonable. It's what you lose when you treat articulability as a proxy for value.
If you want to understand someone's actual value, don't ask them to describe their job. Watch what happens when they're gone.
Every organization has people like this, and you probably have a name in mind right now. Maybe it's the colleague everyone calls when systems are down, not because it's their job but because they somehow always know what to check. Maybe it's the manager whose team never seems to have the drama other teams have. Or the specialist who remembers why that policy exists and what happened the last time someone changed it.
You notice their value only in the absence. You don't notice the crisis prevention until someone stops preventing crises. You don't notice the cultural glue until the team starts falling apart. You don't notice the institutional memory until nobody remembers.
The right question isn't "what do you do" but "what breaks when you leave." And by the time you can answer that question, it's usually too late.
The efficiency consultants always win the argument in the conference room because their questions sound reasonable. "What do you do here?" seems like basic due diligence.
But the question has a design flaw built into its structure. It assumes that all valuable work can be articulated, that inability to explain is evidence of inability to contribute. Polanyi's Paradox says otherwise. The most valuable work often resists articulation precisely because it requires integration, judgment, and context that can't be reduced to bullet points.
The machines are getting better at explicit knowledge. They can summarize documents, answer questions, generate code based on specifications. What remains irreducibly human is the tacit knowledge, the stuff we carry but cannot tell. The intuition that something is wrong before the metrics show it. The judgment about which rules to follow and which to bend. The accumulated sense of how things actually work.
All 76 million boomers will cross retirement age within four years, and most of them can't explain what they do. That's not a sign they're expendable. That's a sign we haven't learned how to value what they know.
Next time: the three kinds of organizational knowledge, and why the one nobody documents is the one that actually matters.