Most of your decisions are two-way doors. You're guarding them like one-way doors.

There’s a framing from Amazon I keep coming back to, because it quietly fixes the most expensive habit small teams have. Some decisions are one-way doors: you walk through, the door locks behind you, and getting back costs more than you’d ever want to pay. Most decisions are two-way doors: you walk through, and if it’s wrong, you walk back out. The mistake almost every early team makes isn’t picking the wrong door. It’s treating every door like it’s the first kind.

Watch where a young product’s time actually goes and you’ll see it. A week of debate over the onboarding copy. A tense meeting about whether the button should say “Start” or “Get started.” A roadmap frozen for a month because nobody can agree on the shape of a settings menu. These are two-way doors — you can change the copy this afternoon, and no user on earth will remember the old verb — but they get guarded with the ceremony you’d reserve for something you can never take back. The cost isn’t the meeting. It’s that while you were deliberating over a reversible thing, the calendar kept moving and the actually-irreversible things went undiscussed.

The move is to make the classification its own step, before any debate about the merits. Ask one question first: if this is wrong, what does it cost to undo? Not “is this important” — importance is a trap, because everything feels important from the inside. Cost-to-undo. If the answer is “an afternoon and a deploy,” you are standing at a two-way door, and the correct behavior is almost rude in its speed: decide now, decide with whoever is closest to the problem, don’t schedule a meeting, and move. You will be wrong sometimes. Being wrong on a two-way door is cheap; that’s the entire point of the door.

The reverse error is rarer but it’s the one that actually hurts you. Some doors are genuinely one-way, and they don’t announce themselves. Your data model is a one-way door — six months of user data written in the wrong shape is not an afternoon to fix. A public pricing promise is a one-way door; you can raise a price, but you can’t un-anger the people who signed up under the old one. A key hire, a platform you build your whole product on top of, the trust you spend the first time you mishandle someone’s data — one-way, all of them, and each is easy to sail through fast because in the moment it feels like just another decision. These are where you want the slow meeting, the second opinion, the night to sleep on it. The skill isn’t being fast or being careful. It’s knowing which is which, and spending your carefulness where it’s scarce.

And here’s the part that turns this from a mental model into an actual build lever: reversibility is often something you design, not something you’re handed. A door you thought was one-way becomes two-way the moment you build the undo. Ship the migration with a rollback path and the schema change gets less terrifying. Put the risky feature behind a flag and launching it is now reversible — you can pull it back in seconds. Run the pricing change as a test on new signups only and you’ve turned a public one-way promise into a quiet two-way experiment. A surprising amount of moving fast safely is just spending a little engineering to convert one-way doors into two-way ones on purpose.

Most of what stalls an early product isn’t hard decisions. It’s easy decisions wearing the costume of hard ones. Sort the doors first. Sprint through the ones that swing both ways, and save your deliberation for the few that lock.