In 2015, Jeff Bezos wrote a letter to Amazon shareholders that has, in the decade since, become one of the most quoted pieces of corporate writing in modern business history.
In it, he proposed a framework that sounds, on first reading, almost too simple to be useful.
He said there are two kinds of decisions.
“Some decisions are consequential and irreversible or nearly irreversible — one-way doors — and these decisions must be made methodically, carefully, slowly, with great deliberation and consultation. If you walk through and don’t like what you see on the other side, you can’t get back to where you were before. We can call these Type 1 decisions.”
Then the pivot.
“But most decisions aren’t like that — they are changeable, reversible — they’re two-way doors. If you’ve made a suboptimal Type 2 decision, you don’t have to live with the consequences for that long. You can reopen the door and go back through. Type 2 decisions can and should be made quickly by high judgment individuals or small groups.”
That’s the framework. Two-way door decisions can be made fast, by small teams, with light process. One-way door decisions get the heavy machinery.
The genius of the framework isn’t in either bucket. It’s in the warning Bezos delivered next.
He said that as organizations get larger, they develop a structural disease. They begin applying the heavy Type 1 process to almost everything — including the vast majority of decisions that are actually Type 2. The result, in his words: slowness, unthoughtful risk aversion, failure to experiment sufficiently, and consequently diminished invention.
Read that last phrase carefully.
Diminished invention.
The act of treating reversible decisions as if they were irreversible doesn’t make a company more careful. It makes the company less inventive. Not by accident. As a direct consequence.
This piece is about why that framework matters more in 2026 than at any moment since Bezos wrote it, and how it explains both the rise of AWS and the failure pattern of most enterprise AI programs I see in the field.
The math nobody talks about
The reason the two-way door framework is so powerful — and the reason most senior leaders systematically underrate it — is that it surfaces a hidden cost that doesn’t show up on any income statement.
The cost of the argument itself.
Consider what happens, in a typical large enterprise, when a Type 2 decision lands on a senior team’s plate. Say it’s a decision about whether to pilot a new tool, restructure a small team, or test a marketing approach.
The process usually looks like this. The originator builds a proposal. A short deck becomes a long deck. The long deck goes through pre-read review with two or three peers. Edits are made. The proposal is scheduled into a leadership meeting. Three weeks of calendar elapse. Five senior people block an hour on their calendars. Someone asks a question that triggers a follow-up analysis. The decision is deferred to the next meeting. Another two weeks of calendar elapse. Eventually a decision is made.
Burn total, conservatively: 3-5 weeks of calendar time, 15-20 hours of senior leadership attention, multiple analyst-weeks of preparation, and an opportunity cost on whatever else those people could have been doing.
The cost of actually running the experiment, for the same decision: usually a week of one person’s time and a small budget that doesn’t need executive approval at all.
The argument is the most expensive form of decision-making humans have ever invented.
The experiment is the cheapest.
For one-way door decisions, the high cost of the argument is justified. The downside of getting an irreversible decision wrong vastly exceeds the cost of the deliberation. You should run the heavy process.
For two-way door decisions, the cost of the argument vastly exceeds the cost of just trying the thing. You should not run the heavy process. You should run the experiment.
This is the core math Bezos was pointing at, and it’s the math most enterprises are systematically refusing to do.
Why this is how AWS services came to life
There’s a common misreading of the Bezos framework that treats Amazon Web Services as the textbook example of a Type 1 decision. The argument: AWS required billions in infrastructure investment, fundamentally redefined Amazon’s business model, and obviously couldn’t be casually unwound.
That reading isn’t wrong about the platform, but it misses the more important story.
AWS is not a single decision. AWS is hundreds of decisions — hundreds of individual services launched, scaled, and (in some cases) quietly shut down over the last two decades. S3. EC2. DynamoDB. Lambda. SageMaker. The full catalog now contains over 200 named services, with new ones launched, modified, or deprecated continuously.
Each one of those services was, at launch, a two-way door.
The first major AWS service, S3 (Simple Storage Service), is documented on Amazon’s own site as having been sketched out by a small team over pints of Hammerhead ale at the Six Arms pub in Seattle’s Capitol Hill neighborhood, and bagels in the Seattle Convention Center. The team didn’t produce a 50-page strategic analysis. They sketched the system, built it, launched it in 2006, and watched what happened.
If S3 had failed, Amazon would have quietly turned it off and walked back through the door. Few outside the team would have noticed.
It didn’t fail.
The same pattern repeated across the AWS catalog. Each service was developed by what Amazon famously called a “two-pizza team” — a team small enough to be fed by two pizzas, with end-to-end ownership of the service, empowered to make decisions without executive sign-off.
Most of the services succeeded. Some didn’t. The ones that didn’t were turned off. Amazon Mechanical Turk, Amazon Polly, Amazon Sumerian — there’s a long list of AWS services that came and went without much fanfare. Two-way doors. Walk back through.
The platform AWS may have been a Type 1 decision at the very beginning, in 2003-2006. The services within AWS — the actual business — were all Type 2 decisions, made quickly by small teams empowered to be wrong.
When you look at it this way, the question stops being “how did Amazon build AWS?” and becomes “how did Amazon build an organization that could launch a new service every other day without senior leadership having to debate it?“
The answer is the two-way door framework, operationalized.
The 2026 amplifier
Bezos wrote the framework in 2015. The structural argument he was making — that arguments cost more than experiments — has been true for as long as organizations have existed.
But there’s something specific about 2026 that makes the framework newly urgent.
The cost of experimentation has collapsed.
In 2015, building a prototype of a new product, workflow, or service took weeks. Setting up the infrastructure, writing the code, gathering data, designing the test — all of it required substantial engineering effort. The math of two-way doors vs. arguments was already favorable, but the gap was meaningful: a multi-thousand-dollar experiment vs. a multi-thousand-dollar argument.
In 2026, you can prototype most software-driven experiments in an afternoon. You can stand up a working AI agent for a specific workflow in hours. You can A/B test a new user experience in a day. The cost of the experiment has dropped to almost nothing.
The cost of the argument has not changed.
A senior leadership debate in 2026 still takes the same three weeks, the same five executives, the same forty slides, the same risk register, that it took in 2015.
The gap between the cost of arguing and the cost of trying has therefore exploded.
This is the single most important structural fact about decision-making in the age of AI, and almost no enterprise has updated its decision metabolism to match.
Most large organizations are running their AI programs through Type 1 processes — vendor evaluations, risk assessments, executive sign-off chains, multi-quarter governance cycles, architecture review boards — because the organization’s default decision metabolism is calibrated for the cost structure of 2015. Maybe 2005.
In that metabolism, the heavy process is justified because the experiments themselves are expensive.
In 2026’s metabolism, the heavy process is not justified, because the experiments are nearly free.
This is why so many enterprise AI programs are failing. Not because the technology doesn’t work. Not because the talent isn’t there. Because the organization’s decision metabolism is producing 18-month argument cycles for decisions that could have been settled by a 2-week experiment.
McKinsey’s 2025 research, widely cited, found that only 5% of companies are capturing substantial AI value at scale. I’d venture that the single largest distinguishing characteristic of the 5% versus the 95% isn’t talent, vendor selection, or budget. It’s whether the organization is willing to make Type 2 decisions at Type 2 speed.
Why senior leaders resist this
If the framework is twelve years old and the math is overwhelming, why does every senior leader I work with revert, instinctively, to applying Type 1 process to Type 2 decisions?
The honest answer is uncomfortable. It’s not that they don’t understand the framework. They understand it perfectly. They resist applying it because their professional identity is bound up with being the careful, deliberate, methodical decision-maker.
For most senior executives, the path to senior leadership rewarded heavy process. The thoughtful analysis. The thorough deck. The well-considered judgment. The careful synthesis of input from multiple stakeholders. That was the skillset that produced promotions for two or three decades.
The two-way door framework asks those executives to let go of the very skill that got them where they are, for the majority of the decisions on their plate.
That’s not a process issue. That’s an identity issue.
I see this every week. A senior leader will agree, in principle, that a particular decision is a two-way door. They will then proceed to organize three meetings, produce a 30-slide deck, and pull in five peers — for the two-way door decision they just agreed was reversible.
The behavior isn’t irrational. The behavior is identity-preserving. The heavy process is what makes the executive feel like an executive.
The work of moving an organization to operate at Type 2 speed for Type 2 decisions is therefore not primarily a process redesign. It’s an executive coaching exercise. Helping senior leaders separate their self-image from the volume of deliberation they apply to small decisions.
This is one of the most important — and most rarely addressed — leadership development conversations of the next decade.
Three practical moves
If you want to start operationalizing the framework in your own organization, three places to start.
One: audit your last ten decisions.
Take the last ten substantive decisions you personally made or were involved in. Honestly classify each one. Type 1 or Type 2? Reversible or not?
In every enterprise audit I’ve done with senior teams, the result is roughly the same. Three or four decisions out of ten that the leadership team treated as Type 1 were, on honest examination, Type 2. Reversible. Small downside if wrong. They got the heavy process anyway.
That ratio is the gap between how fast your organization is moving and how fast it could be moving.
Two: force the question into the room.
Before any debate begins, ask one question explicitly: “Is this a one-way door or a two-way door? If it’s a two-way door, why are we arguing about it instead of testing it?”
The question alone, asked consistently for a few weeks, will retrain the room. People start preempting it. They stop bringing two-way door decisions into long-form deliberation. The decision metabolism speeds up.
Three: build a two-way door budget.
Allocate a small pre-approved pool — $25,000, $50,000, whatever your scale justifies — that any individual or small team can deploy on a Type 2 experiment without executive sign-off.
The dollar amount is not the point. The signal is the point. The very existence of the budget tells the organization: we have stopped treating small reversible decisions as if they were big irreversible ones.
This single structural change, in my experience, does more to shift an organization’s invention rate than any number of innovation workshops, transformation initiatives, or culture decks.
The closing thought
There’s a sentence Bezos wrote, in passing, in the 2016 follow-up to the original two-way door letter, that has stayed with me as the cleanest articulation of the entire executive challenge in the age of AI.
“If you’re good at course correcting, being wrong may be less costly than you think, whereas being slow is going to be expensive for sure.”
Read it slowly.
Being wrong may be less costly than you think.
Being slow is going to be expensive for sure.
The default mode of senior executive decision-making in most large organizations is optimized for not being wrong. That mode was calibrated for a world in which experiments were expensive and arguments were the only cost-effective way to filter ideas.
That world is over.
In the world we now live in, experiments are nearly free and arguments are nearly the only thing that’s still expensive. The optimization function has flipped. Being wrong, recoverably, on a Type 2 decision is now a competitive advantage — because it means you’re moving. Being slow, while you make sure you’re not wrong, is the new failure mode.
Bezos saw this twelve years ago. He named it. He built an organization around it. He turned a 1990s online bookstore into the company that, on the back of two-way door experimentation, became one of the largest enterprises in modern economic history.
The framework is sitting there, free, for anyone to apply.
The hard part isn’t understanding it.
The hard part is being willing to let go of the kind of executive you used to be in order to become the kind your organization now needs.
That’s the only real question.
Try it. It’s cheaper than arguing about it.
The world has changed. The leaders who notice will be the ones the next decade is built around.
