Disagree and Commit

Decisions must be made under disagreement, and the alternatives to a clean resolution are all corrosive: relitigating continuously, executing half-heartedly, or escalating everything. Disagree and commit is the mechanism that lets an organisation move while preserving the information that the disagreement contained.

Its two halves are equally important and the second is the one that is skipped. Recording the dissent without committing is obstruction; committing without recording loses the learning.

Why it matters

A half-executed approach fails for reasons unrelated to the decision, which proves nothing, harms the organisation, and destroys the dissenter's credibility for the next disagreement — because the failure is attributable to the execution rather than to the choice.

The recorded trigger is what makes the practice more than a courtesy. A dissent stated as "I expect this to fail when we reach X, because Y" is a monitorable position, and when X arrives the organisation revisits the decision on evidence rather than on who is more insistent. Stated as a general objection, it is unfalsifiable and provides nothing.

Implementation patterns

• Establish first whether the disagreement is about objectives. "What are you optimising for that I might not be weighting properly?" frequently resolves it entirely, by revealing a constraint that makes the other position correct.
• Separate reversible from irreversible. For a reversible decision, try it and gather evidence — arguing costs more than testing. Reserve escalation for decisions that are expensive to undo, and say explicitly that this is why you are pressing.
• State the position as a prediction with a condition, which can be tested.
• Propose the cheapest experiment that would settle it — a load test, a spike, a prototype of the risky part. Converting an argument into an experiment is the single most effective move available, because it removes the need for either party to be persuaded.
• Name the evidence that would change your mind, which is credible and tends to be reciprocated.
• Record: the decision, the reasoning, the dissent, and the revisit trigger, stated neutrally rather than as a prediction of vindication.
• Then commit visibly and completely, including advocating for the approach with the team.
• Report the outcome afterwards, including when you were wrong — the only durable source of influence available.

Industry example

The phrase is most associated with Amazon's leadership principles, where it is paired explicitly with the obligation to disagree first: the practice fails if the disagreement is not genuinely voiced, and it becomes a euphemism for compliance.

The underlying idea is broader and appears in every effective architecture practice: decision records that include dissenting views and revisit conditions, review forums that record accepted risks alongside decisions, and the general principle that an organisation's ability to learn depends on having written down what it expected before finding out.

Failure scenarios

• Commit without genuine disagreement, where the phrase becomes a way to close discussion.
• Disagreement without commitment, executed half-heartedly, invalidating the experiment.
• No recorded trigger, so the decision is revisited on assertion rather than evidence — or not at all.
• A trigger written as a prediction of vindication, which is read as positioning and ignored.
• Escalating reversible decisions, which spends credibility on things that could have been tested.
• Escalating without presenting the other position fairly — an escalation that misrepresents the other side is resolved against you regardless of technical merit.
• Never reporting outcomes, so the record is written and never closed.
• Dissenting on everything, after which genuine concerns are discounted with the rest.

Trade-offs

The practice requires the dissenter to execute something they believe is wrong, which is genuinely uncomfortable and occasionally means contributing to a foreseeable failure. That is the cost of an organisation that can move, and it is real.

It also can be used to suppress disagreement — "disagree and commit" invoked as a way to end a conversation that had not happened. The obligation to disagree first is what prevents this, and where the culture does not support voicing dissent, the practice degrades into compliance with better branding.

The trade is individual discomfort and occasional foreseeable failure in exchange for decisiveness plus a learning record. The alternative — endless relitigation or unaccountable half-execution — is worse on every dimension, which is why the practice is close to universal among organisations that ship.

Interview question

"You have argued against a decision, you have lost, and you still think it is wrong. Tell me exactly what you write down, what you say to your team the next morning, and what would have to happen for you to raise it again."