Your team has debated two CI setups for five weeks. What do you do?

Classify it as a two-way door (switching costs days), let the owner decide now, record the minority concern as a revisit trigger (e.g. build time over 15 minutes), and ask everyone to disagree and commit.

Give two examples of one-way-door decisions in software and one way to make each more reversible.

Choosing a primary data store (keep a repository layer and plan dual writes for migrations); publishing a public API contract (version it and deprecate gradually); signing a long vendor contract (abstraction layer, shorter term).

What makes 'disagree and commit' healthy rather than a way to shut people up?

The disagreement is genuinely heard and written down first, the decision owner explains the reasoning, and there's a revisit trigger. Then everyone commits fully, including those who disagreed.

Decision Speed: One-Way and Two-Way Doors — The Tech Lead Handbook

Scenario

Example team: the platform team needs a new CI runner setup. The discussion between two options has lasted five weeks, with benchmark after benchmark. Meanwhile the same team picked a primary database for a new service in a 30-minute meeting.

What you see on the team

Small, easily changed decisions (a library, a CI setup, a folder structure) debated for weeks.
Large, hard-to-change decisions (data stores, public APIs, data models) made quickly by whoever is in the room.
Meetings that end with "let's gather more data".
Frustration from both the cautious and the fast-moving engineers.

The common mistake

Using one decision process for everything — either everything is slow and careful, or everything is fast and casual.

Why it happens

Teams don't classify decisions by reversibility. Debates about small choices are comfortable and visible; big choices are often made implicitly, as a side effect of a project. Amazon's 2016 shareholder letter puts it simply: "Many decisions are reversible, two-way doors. Those decisions can use a light-weight process." The same letter suggests most decisions "should probably be made with somewhere around 70% of the information you wish you had".

The playbook

Classify first — ask "what does it cost to undo this in six months?"

	Two-way door (reversible)	One-way door (hard to reverse)
Examples	CI tooling, internal library, code layout, feature-flagged behaviour	Primary data store, public API contract, data model with migrations, vendor lock-in with long contract
Who decides	The engineer or pair doing the work	Tech lead with review; ADR
Process	Short write-up, try it, measure	Design doc, alternatives, review, ADR
Time box	Hours to days	Days to a few weeks

Make one-way doors more reversible where cheap: abstraction around a vendor, feature flags, dual writes during migrations, versioned APIs.

End stalemates explicitly. When arguments are repeating and the door is two-way, the decision owner decides. The same letter recommends the phrase "disagree and commit": you can disagree and still commit fully to the chosen path.

Write a revisit trigger: "We'll revisit if build times exceed 15 minutes" — so disagreement has a constructive outlet.

What to say

Classifying in a meeting:

"Before we go further — is this a one-way or two-way door? If we pick the wrong CI runner, what does it cost to switch in three months? A week? Then let's decide today, try it for a month, and revisit if build times go above 15 minutes."

Closing a stalemate:

"We've heard the arguments twice. I'm going with option B for the reasons in the doc. Tom, I know you'd have chosen A — can you disagree and commit? I've written down your concern as the revisit trigger."

Trade-offs

Fast decisions on two-way doors occasionally cost a rework; that is usually cheaper than weeks of debate.
"Disagree and commit" can be abused to silence dissent; it only works if the disagreement is heard and recorded first.
Reversibility is a spectrum; when unsure, ask what data, contracts or users would be affected by a reversal.

Signals it's working

Reversible decisions take days, not weeks.
Irreversible decisions have a design doc and ADR.
Disagreements are recorded with revisit triggers, and some are actually revisited.
People commit to decisions they argued against.

Key points

Classify decisions by the cost of reversing them before choosing a process.
Two-way doors: decide fast, close to the work, with a revisit trigger.
One-way doors: design doc, alternatives, review, ADR — and make them more reversible where cheap.
End stalemates explicitly: hear and record the disagreement, then disagree and commit.