One-Way and Two-Way Doors
A one-way door is a decision that is very hard or very expensive to reverse. A two-way door is one you can walk back through if it turns out to be wrong. Jeff Bezos popularised the idea, calling them Type 1 and Type 2 decisions: the trick is to match the speed and rigour of the decision to how reversible it is.
The Core Idea
- Two-way doors: Decide fast, with whoever is closest to the problem, using roughly 70% of the information you wish you had. If it goes badly, you walk back and lose little.
- One-way doors: Slow down. Gather evidence, write it down, involve the right people, and make the call deliberately.
Most organisations get this wrong in one of two directions. They treat everything like a one-way door (committees, sign-offs, endless debate), or they treat everything like a two-way door and discover too late that something was permanent.
Telling the Doors Apart
Ask three questions:
- Cost to undo: How much time, money, and goodwill would reversing this take?
- Blast radius: How many teams, customers, or systems are affected while it is live?
- Lock-in: Does this create data, contractual, or skills dependencies that outlast the decision?
If the answers are "low, small, none", walk through. If any of them is "very high", treat it as a one-way door.
Examples for a CTO
| Usually two-way | Usually one-way |
|---|---|
| Internal tooling or library choice | Primary cloud provider with deep proprietary service use |
| Feature behind a feature flag | Public API contract or data schema customers depend on |
| Team rituals and meeting cadence | Selling or open-sourcing core IP |
| Trying a new CI pipeline | Large multi-year vendor contracts |
| Pilot with one team | Rewriting the core platform from scratch |
| Hiring on a trial or contract basis | Senior leadership hires and reorganisations (reversible on paper, costly in trust) |
Doors Change Over Time
You can often turn a one-way door into a two-way door with a little design work:
- Put an abstraction or anti-corruption layer in front of a vendor.
- Roll out behind feature flags, canaries, or blue/green deployments.
- Run a time-boxed pilot before committing to the full rollout.
- Negotiate shorter contract terms or exit clauses.
The inverse also happens. A harmless shortcut becomes a one-way door once other teams depend on it. Revisit your classification when adoption grows.
Why CTOs Should Care
- Speed: Pushing two-way decisions down the org chart removes bottlenecks and builds ownership.
- Safety: Reserving heavy process for one-way doors means rigour is spent where it matters.
- Less paralysis: A shared vocabulary ("is this a two-way door?") ends many debates in seconds. See Analysis Paralysis.
- Better records: One-way doors are exactly the decisions worth capturing in an ADR.
Common Pitfalls
- Everything is a one-way door: Fear and low trust make every choice feel permanent. Fix the safety net, not the approval chain.
- Disguised one-way doors: Quick experiments that quietly become production dependencies.
- Ignoring the cost of staying: Not deciding is also a door, and it often closes by itself.
- Reversible in theory only: If rolling back needs a heroic migration nobody has rehearsed, it is a one-way door.
Practical Takeaway
Before any significant decision, say out loud: "Is this a one-way or a two-way door?" Then pick the process to match. Two-way: decide today, review in a sprint. One-way: write the options, name the owner, set a decision date, and record the outcome.
Explore Next
- Analysis Paralysis — What happens when every decision is treated as irreversible.
- Architectural Decision Record (ADR) — Write down the one-way doors so future you knows why.
- ROAM Risk Management — Resolve, Own, Accept or Mitigate the risks a risky door brings.
- OODA Loop — Faster decision cycles are only safe when mistakes are cheap to undo.
References
- 2015 Letter to Shareholders, Jeff Bezos (Amazon) — Where Bezos describes Type 1 and Type 2 decisions.
- Reversible Decisions (Wikipedia, Decision-making)Wikipedia — Background on decision-making models and biases.