Working Backwards

The book's organising claim is that stated principles decay unless they are embedded in a repeatable process with a tool, an owner and an inspection cadence. Amazon calls these mechanisms, and the authors argue they are what actually transmits culture at scale.

Their formulation — good intentions do not work, mechanisms do — is the thesis. Everything else in the book is an example of a mechanism.

Good intentions do not survive scale. A mechanism has a tool, an owner and an inspection, and survives.

Devmark’s reading — not a quotation from the book

The practice that gives the book its title: before building anything, write the press release announcing it and the frequently-asked-questions document answering the hard objections.

The discipline is that the document must be written in customer language, must state the benefit plainly, and must survive a room full of people trying to find the flaw. If the press release is not compelling, the product is not worth building — and no code has yet been written.

Amazon replaced presentations with six-page prose documents read silently at the start of a meeting. The authors are emphatic about why: slides let a presenter conceal weak reasoning behind a confident delivery, while prose forces the logic into the open.

The silent reading is the part most imitators skip and the part that makes it work — it guarantees everyone has actually engaged with the argument before anyone speaks.

Their hiring mechanism: a trained interviewer from outside the hiring team with veto power over the decision, whose interest is the long-term quality of the company rather than filling the role.

The structural insight is that a hiring manager under pressure has an incentive to lower the bar, and that the only reliable counterweight is someone with no stake in the vacancy and the authority to say no.

Their answer to why initiatives inside large companies stall: the people responsible have other jobs. A single-threaded leader owns one thing and nothing else.

This is paired with separable, single-threaded teams whose dependencies have been deliberately removed — the organisational reasoning that produced the service-oriented architecture, which the authors present as an organisational decision that happened to have technical consequences.

The architecture followed the org chart on purpose. Removing dependencies between teams was the objective.

Devmark’s reading — not a quotation from the book

Their metrics discipline: output metrics like revenue tell you what happened; input metrics like selection, price and convenience are what you can actually act on.

The weekly business review inspects controllable inputs rather than celebrating outputs, on the reasoning that the outputs follow if the inputs are right — which is the same logic as Grove's leading indicators, applied at considerable scale.

Extended accounts of Kindle, Prime, Amazon Prime Video and AWS, told as demonstrations of the mechanisms rather than as triumph narratives.

The Kindle account is the most instructive because it includes the decision to build hardware against the company's own competence, and the authors are reasonably candid about the internal resistance and the mistakes.

This needs stating. The authors write about mechanisms and largely do not engage with the labour practices, warehouse conditions, seller relationships or competition questions that define much of the public argument about the company.

They are insiders writing about what they built, and the omission is structural rather than incidental. A reader should get that material elsewhere — Brad Stone's reporting, for instance — rather than expecting it here.

The narrative memo and working-backwards document transfer well to organisations of almost any size and are the most commonly adopted.

Bar Raisers and single-threaded leadership require organisational scale and authority to implement, and attempts to adopt them in small companies usually produce ceremony without the substance. The authors do not much distinguish which mechanisms depend on size.

Write the press release first. It is free, it can be done this week, and it reliably kills projects that should not proceed before anyone has spent money on them.

Then the input-metrics discipline: list what you actually control, and check whether your review meeting inspects those or only reports the outcomes they eventually produce.