Escalation Is a Service, Not a Betrayal

Escalation is a routing mechanism: it moves a decision to the smallest scope that has the authority and context to make it. Refusing to escalate doesn't resolve a conflict — it resolves it by exhaustion.

The misconception

Engineers carry three false models of escalation, usually all at once:

Escalation as snitching. Taking a disagreement to managers feels like reporting a peer to the teacher. It violates the engineer's self-image of solving problems directly, peer to peer.
Escalation as personal failure. "I should be able to resolve this myself." Involving a manager is an admission that you couldn't.
Escalation as the nuclear option. Something you do once, at the bitter end, when the relationship is already burned — so you save it, and save it, and never use it.

The result is predictable: an engineer grinds for six weeks in a stuck cross-team disagreement — re-explaining the same position in meeting after meeting, writing longer and longer docs, hoping the other side will eventually just agree — when a 15-minute conversation between two managers would have settled it in week one. The weeks of grinding weren't diligence. They were a routing failure.

The reframe: escalation is routing

Every decision in an organization has a smallest scope that can legitimately make it — the lowest node in the org tree that has both the authority and the context to decide. Escalation is nothing more than moving a decision to that node. That's it. It's not a moral act, it's an addressing scheme.

This reframe makes one thing obvious that the grinding engineer misses: two peers deadlocked on priorities literally cannot resolve the deadlock. Not "it's hard" — structurally cannot. If Team A needs Team B to prioritize an API change, and Team B's roadmap says otherwise, neither engineer has authority over the other's roadmap. The decision they're fighting over does not live at their level. More meetings between them cannot produce a resolution, because no combination of arguments changes who owns the tradeoff. The common manager owns it. Routing it there isn't giving up on the peer relationship; it's delivering the decision to its actual owner.

What happens if you refuse to route it? The conflict still gets resolved — by stamina. Whoever can attend more meetings, write more rebuttals, and outlast the other wins. This is the point Bezos made in the 2016 shareholder letter: "disagree and commit" and immediate escalation exist precisely because the alternative is exhaustion-based decision-making, which is slow, demoralizing, and selects for persistence rather than correctness. A stamina contest produces a decision too — just a worse one, later, at higher cost, decided by the wrong variable.

Seen this way, escalation is a service you perform for the organization: you're the person closest to the deadlock, so you're the one who can see it's stuck and route it upward. Sitting on it is withholding information from the people who own the tradeoff.

The cost curve

Escalation cost is not constant — it compounds with delay.

Day 2: "We've found a genuine priority conflict, can you two align?" Cost: one meeting. No blame available to assign; nothing has been lost yet. This is a routine routing event.
Week 3: Some schedule damage, some accumulated frustration, a few meetings' worth of positions hardened. Cost: a slightly tense alignment conversation plus rework.
Week 6: The deadline is blown or about to be. Both teams have spent goodwill fighting. And now the escalation itself is contaminated: the first question from every manager is "why am I hearing about this now?" You pay for the deadlock and for the delay in reporting it.

The perverse psychology: the longer you wait, the more the escalation feels like an admission of failure — so the more you avoid it — so the worse it gets. Early escalation is cheap precisely because there's no failure to admit yet. The move is to escalate while it's still boring.

A manager who learns about a six-week deadlock at deadline time has been failed twice: once by the deadlock, once by not being told. Managers consistently prefer early escalation to late surprises — routing decisions to their level is a core part of what their job is. An early escalation is you doing their intake for them; a late one is you handing them a crisis.

How to escalate well

Warn before escalating. "I think we're stuck — shall we take this to our managers together?" Never escalate as an ambush. This sentence alone often unsticks the conflict: it signals you're serious, it's not a bluff, and it gives the other side one clean chance to move. If they move, you never escalate. If they don't, they've agreed to the escalation.
Escalate jointly. The trust-preserving form is "we disagree and need a call" — framed together, presenting both positions fairly, ideally in one shared doc or one meeting with both of you in the room. Joint escalation carries no blame; it's two professionals reporting a structural conflict. Unilateral escalation behind someone's back is what actually earns the "snitch" label.
Escalate the decision, not the person. Frame: "customer need X is unmet; we need a call on whether it beats priority Y." Never: "Team B is being difficult." The first is an unarguable statement of a tradeoff that needs an owner. The second starts a turf war, forces the other team to defend itself, and makes the deciding manager adjudicate a personality dispute instead of a priority call.
Present the other side's position well. If the other team would sign off on your summary of their argument, your escalation is trustworthy. Strawmanning them buys you one win and a reputation.
Come with a decision request, not a vent. "We need a call between A and B by Friday; here are the two options and their costs" — a routable packet. "This has been so frustrating" is not.
Keep your own manager in the loop before it's formal. A heads-up in a 1:1 ("this might need escalating soon") costs nothing and means the escalation, when it comes, is expected.

Failure modes

Escalating everything. If every disagreement goes upstairs, you're not routing — you're outsourcing your job. It spends your capital, trains managers to discount your escalations, and signals you can't operate at your level. The filter: escalate only when the decision genuinely lives above you (authority mismatch), not when you merely haven't persuaded someone yet. Exhaust the direct conversation once, properly, first — then route.
Escalating around your own manager. Skipping levels, or letting your manager hear about the escalation from their boss, converts a routine routing event into a surprise about you. Whatever the merits, the story becomes "why didn't they come to me?" Your manager should never learn about your escalation from above.
Escalating as a weapon. Using the threat of escalation to bully peers into compliance ("agree or I'm going to your boss") wins the issue and poisons every future interaction. The warn-first move works because it's an invitation to escalate together, not a threat.
Escalating without a position. Dumping a raw conflict upward forces the manager to reconstruct everything. Route the decision with the context attached: options, costs, your recommendation, the other side's view.

The diagnostic test

When you're stuck with another team, ask: does anyone in this conversation have the authority to make the decision we're fighting about?

If yes — keep talking; you haven't finished the direct conversation. If no — every additional meeting at this level is a stamina contest wearing a collaboration costume. Route it. Today it costs one meeting; in a month it costs the deadline, the goodwill, and the question "why didn't you raise this earlier?"