Skip to content
Latticework
Esc
Try
No. 007Ways of Thinking

One-Way vs Two-Way Doors

Match decision rigour to reversibility: decide reversible choices fast and irreversible ones carefully.

“Some decisions are consequential and irreversible or nearly irreversible – one-way doors – and these decisions must be made methodically, carefully, slowly…”

Spend your rigour on the decisions you cannot walk back, and move fast on the ones you can.

To an executive

Treating every decision as high stakes slows the organisation and stifles experiments. Reserving heavy scrutiny for choices that cannot be undone speeds things up without adding risk.

To an engineer

Classify each decision by how hard it is to reverse. Decide reversible ones quickly, deliberate over irreversible ones, and try to make hard-to-reverse choices easier to undo.

The idea

In his 2015 letter to Amazon shareholders, Jeff Bezos described "one-size-fits-all decision making" as a common pitfall of large organisations. Some decisions are consequential and irreversible or nearly so: one-way doors, which he called Type 1. They deserve slow, careful deliberation. Most decisions are two-way doors, Type 2: if you get one wrong, you can go back through. Those should be made quickly by individuals or small groups with good judgement.

The trouble, he wrote, is that as organisations grow they tend to use the heavy Type 1 process for most decisions, including Type 2 ones. The result is slowness, too little experimentation and less invention.

The architect's job is often to classify decisions correctly, and sometimes to turn a one-way door into a two-way door.

Why it works

Rigour is a limited budget. Review boards, prototypes and long analysis are expensive. Spending them on reversible choices leaves less for the ones that matter.

Reversible choices are cheaper to learn from. Trying something and changing it later often teaches more than debating it up front.

Reversibility can be designed. Abstractions, feature flags, staged rollouts and versioned contracts all lower the cost of going back, which moves decisions from Type 1 to Type 2.

Where it applies

  • Usually two-way: front-end frameworks, internal libraries, most configuration, tooling and the internal structure of a single service.
  • Usually one-way: primary data stores and data models, identity providers, public API contracts, data residency and long vendor contracts.
  • Vendor selection, where exit costs decide which kind of door it is.
  • Governance design, where review effort should follow reversibility rather than size or visibility.

How to apply

  1. Ask what reversal would cost in time, money and disruption, not just whether it is possible.
  2. Look for what makes it one-way: accumulated data, external consumers, contracts, migrations, user habits.
  3. Make it reversible where you can: put an interface in front, flag it, pilot it with one team or keep the old path running.
  4. Match the process to the door. Write an ADR, prototype and review one-way decisions; let teams decide two-way ones and record them briefly.

Example scenario

A product team spent six weeks evaluating logging libraries, with comparison matrices and a review board slot. In the same quarter, the multi-tenant data model was chosen in an afternoon workshop: one shared schema with a tenant column.

Two years later, a large customer required its data to be held separately. Swapping the logging library would have taken a week; splitting the tenancy model took most of a year. The architecture team then introduced a short reversibility check into design intake, so data-model and contract decisions got the deep review and library choices were left to teams.

Questions to ask

  • What would it cost to reverse this in twelve months, once data and users depend on it?
  • Can we make this reversible with an abstraction, a flag or a contract?
  • Are we over-deliberating something cheap to change?

When not to use it

  • Decisions that are cheap either way, where even classifying them is more effort than the choice deserves.
  • Safety and security incidents, where speed matters more than deliberation and the immediate action is usually prescribed.

How it misleads you

It is tempting to label everything a two-way door so you can move fast. Many "reversible" choices become one-way once data, contracts or customers build up on them. Reversibility is not fixed: a door that is two-way on day one may be one-way a year later, so revisit the classification as a system matures.

In the wild

Further reading

Last reviewed · Suggest an improvement