When "Fail Fast" Is Malpractice

We were three days into a MedTech sprint, the energy was good, and I was doing what I do in every sprint: pushing the team to stop talking and start shipping. It's the instinct I spend most of my time building in teams, and most days it is exactly the right one. Build the rough version, get it in front of users, learn fast. Fail fast.

Then someone — an engineer who'd been quiet all morning — asked the question that mattered: "Fail fast on which part, exactly?" Because the part we were about to ship fast touched a dosing calculation. And there is no fast, cheap, reversible way to be wrong about a dose.

That's the moment "fail fast" cracked open for me. Not as wrong — as half a sentence. The gospel of modern innovation — move fast, break things, ship and learn — is the best advice in the world, right up until the thing you'd break is a person, a balance sheet, or a number nobody can take back. And nobody ever tells you where that line is.

So let me try. Here are two decisions. Ship each one:

You felt it in your hand. The button colour you flipped back without a thought. The dose you couldn't take back, no matter how many times you clicked. Same instinct — "just ship it" — and one is a learning loop while the other is a liability. The difference was never speed, or courage. It was reversibility.

Fail fast was always about two-way doors

The clearest language for this comes from Jeff Bezos, who split every decision into two kinds. A two-way door is reversible: walk through, look around, and if you don't like it, walk back out. A button colour, a landing-page headline, a meeting format. A one-way door is not: once you're through, you live with it. He argued that reversible decisions should be made fast, by small groups, with barely any process — and irreversible ones slowly, carefully, with real deliberation.

"Fail fast" is simply two-way-door advice that forgot to mention the door. In a startup shipping a consumer app, almost every door is two-way, so the advice feels universal. Carry it into MedTech, aviation, payments, manufacturing — the places I keep getting invited precisely because the stakes are high — and suddenly half the doors are one-way. Run the consumer-app playbook there, and "break things" stops being a motto and starts being an incident report.

Here's the tell that the gospel has limits: even its author retired it. "Move fast and break things" was Facebook's motto until 2014 — when Facebook, having grown big enough that breaking things meant breaking people's things, quietly changed it to "Move fast with stable infrastructure." Same company, same love of speed. They'd just learned which doors they were standing in front of.

The fix is not "slow down"

This is where most risk-averse organisations get it exactly as wrong, only in the opposite direction. They look at one genuinely irreversible decision, conclude the whole domain is dangerous, and wrap everything in the heavyweight, one-way-door process: every change reviewed to death, every experiment a project, every prototype a six-month committee. Bezos's warning cuts both ways — treating two-way doors like one-way doors is its own slow, expensive failure. It just doesn't show up as an incident, so nobody counts it.

The actual skill — and the thing the best regulated-industry teams do that the cautious ones don't — is to find the two-way doors inside a one-way problem. You almost never have to bet the irreversible thing in order to learn. You can:

• Move the experiment off the live system. A simulation, a sandbox, a shadow mode running beside the real thing but wired to nothing — these turn a one-way decision into a two-way rehearsal. The clinicians can "use" the new dosing tool on synthetic patients all week; nobody's at risk, and you still learn fast.
• Shrink the blast radius until it's reversible. A one-way door for the whole hospital is a two-way door for one ward, for two weeks, with a human checking every output. Pilots, ramps, reversible rollouts — that's how you make an irreversible change survivable enough to fail fast on.
• Separate the reversible surface from the irreversible core. Most "high-stakes" projects are 80% interface, workflow and wording — all two-way — wrapped around a 20% core that genuinely can't be wrong. Iterate like a maniac on the 80%. Treat the 20% like the one-way door it is.

Done well, you keep the entire learning engine of fail-fast — the speed, the prototypes, the real user contact — and aim it only at the doors you can walk back through. That's not a compromise between speed and safety. It's what fast actually looks like when the stakes are real.

The only question that matters

"Should we move fast or be careful?" is the wrong question, and it produces the two classic failures: cowboys shipping irreversible harm, and committees reviewing reversible trivia for a quarter. The right question is smaller, and answerable, and you can ask it of any decision on the table: is this a two-way door?

If yes: stop deliberating and go — that's the bet you can afford to lose, and sizing exactly that kind of bet is what I built a whole tool around. If no: slow down, give it the deliberation a one-way door deserves, and then go hunting for the two-way door hidden inside it. (It's a close cousin of reading the moment — except the variable isn't time, it's whether you can take it back.)

"Fail fast" isn't wrong. It's just the second half of a sentence. The first half is: through doors you can walk back through.

Sizing a bet — and not sure whether it's a door you can walk back through? That's exactly what The Bet is for. Or let's talk through a decision you can't quite call.