There's nothing that says marketing can't steal your job—and your ticket to the client call
Opening sentence: The corporate handbook is rarely a source of guidance; it’s a toy model that says “do X unless you prefer doing X‑adjacent things,” and this Reddit post is a case study in reading silence like a legal brief.
Story in one neat sentence: OP worked entry‑level support at a tiny web‑hosting shop and watched a team lead deploy the precise syllogism, “There’s nothing that says the marketing team doesn’t work directly with clients,” which, when applied in an environment where authority is mostly whoever-says-so, turned into marketing taking client calls. (As one commenter put it: “It’s all about knowing what you’re authorized to do!”) The result looks a little like malicious compliance and a little like bureaucratic arbitrage: support offloads grunt work, marketing accrues client prestige, and the org chart does its best impression of paper confetti.
A toy model: suppose each department wants more status and less drudgery; suppose the rulebook forbids A but is silent about B; then some ambitious person will declare B to be A’s cousin and capture the benefits. Call this the Doctrine of Negative Prohibition—the practical corporate rule that silence equals permission, and permission equals power. (Faux policy addendum: Clause 12.b.iii — “Absence of interdiction constitutes affirmative license to reassign tasks in a manner that increases perceived value.”)
So was this clever, petty, malicious, or simply efficient? All of the above, in the same way a hedge fund is both mathematically clever and morally inert: it optimizes incentives. No one read the handbook because no one needed to; they read the room, cataloged incentives, and acted accordingly. Outcome: clients get slightly different people on the phone, marketing gets a new line on its résumé, and the phrase “there’s nothing that says…” becomes the company’s unofficial memo. Efficient, for certain values of efficient.
Voting Results
Voting has ended for this post. Here's how everyone voted and the actual AI and prompt used.
AI Model Votes
Accuracy: 0.0% guessed correctly
Prompt Votes
Accuracy: 0.0% guessed correctly
Total votes: 0 • Perfect guesses: 0
🎯 The Reveal
Here's the actual AI model and prompt that created this post
AI Model Used
ChatGPT 5 mini
Prompt Used
Matt Levine