पाठशाला Pathshala · मन Man, The founder · Lesson 04 · Start

Decisions: reversible, irreversible and how fast to make each

Most decisions can be undone and should be made in an afternoon. A few cannot and deserve a memo. The founder’s skill is telling them apart, and most companies get it backwards.

Pathshala, The Founder Library · 11 October 2026 · 8 min read

A founder makes several dozen decisions a week and treats nearly all of them the same way, which is to say with whatever energy is left. The result is the worst of both: the small ones take meetings and the large ones get made at eleven at night because the meetings used up the day. The fix is not better judgement. It is sorting, before deciding, into the few that lock and the many that do not.

This lesson gives the sort in two questions, the speed each category deserves, and the one-page memo for the decisions that earn one. The tree in the middle is the whole method; the sections around it are why each branch goes where it does.

Two kinds of door

Jeff Bezos put the distinction into his 2015 letter to Amazon shareholders and it has not been improved on since. Some decisions are consequential and irreversible or nearly so; he calls these one-way doors, Type 1, and says they must be made methodically, carefully, slowly, with great deliberation and consultation. But most decisions are not like that. They are changeable, two-way doors, Type 2, and if you have made a suboptimal one you do not have to live with it for long. Type 2 decisions, he writes, can and should be made quickly by high-judgement individuals or small groups.

The passage founders should read twice is the warning. As organisations get larger there is a tendency to use the heavyweight Type 1 process on most decisions, including many Type 2 ones, and the end result is slowness, unthoughtful risk aversion, failure to experiment sufficiently and diminished invention. Bezos was writing about a company with hundreds of thousands of employees. The same drift happens in a company of four, and it starts the week a founder calls a meeting to pick a project-management tool.

The test for which door you are at is a single question: if this turns out to be wrong, can it be reversed within a month without lasting damage? A price shown to the next twenty customers, a tool, a landing page, a channel experiment, a feature behind a flag: yes. A hire, a lease, a public promise to a customer: mostly, but it leaves a mark on someone. A co-founder, an institutional equity round, the market you choose to serve, a contract with a lock-in, a name on a registration that customers will learn: no. The glossary has the [type 1 and type 2 entry](/library/glossary) if the words are new.

The second axis: what it costs to be wrong

Reversibility alone is not enough, because a reversible decision can still be expensive. Spending three months of the whole team on a feature that can be deleted is reversible in the strict sense and ruinous in the practical one. So the second question is cost, in the two currencies a small company actually has: the team’s attention and the bank balance. A useful threshold is two weeks of the team’s time or one month of burn. Under both, the decision is cheap enough to be wrong about. Over either, it deserves a paragraph.

The two questions make four boxes. Reversible and cheap is most of the week: one person decides today, no meeting. Reversible and costly gets a paragraph and a week, and the paragraph exists so that reversing it later is a decision rather than a drift. Irreversible and cheap is rare and usually turns out, on inspection, to be irreversible for someone else: a small promise, a small hire. Treat it as the next box. Irreversible and costly is the handful of decisions a year that shape the company, and they get the memo described below. Companies that fail at decisions usually fail by putting the first box through the fourth box’s process and the fourth through the first’s.

How fast: the seventy per cent rule

The following year Bezos added the speed. In the 2016 letter, under the heading of high-velocity decision making, he writes that most decisions should probably be made with somewhere around seventy per cent of the information you wish you had, and that if you wait for ninety per cent you are in most cases probably being slow. The reason it works is the first axis: for a two-way door, being wrong is cheaper than being slow, and the remaining thirty per cent of the information usually arrives faster by deciding than by waiting for it. Speed, he says plainly, matters in business.

Two more tools from the same letter apply directly to a founding team. The first is the phrase disagree and commit: if you have conviction on a direction even though there is no consensus, ask the others to gamble with you on it. It is not a demand for silence. It is a way to end a debate that has been had, with the person who lost it still owning the outcome. The second is to recognise true misalignment early and escalate it immediately, because the alternative is the decision being settled by whoever has the most stamina, which in a company of three means the decision is made by exhaustion.

For a reversible decision, then, the clock is a day if it is cheap and a week if it is not, and the only acceptable reason to extend the week is a test that can be run inside it: five customer calls, a price shown to the next cohort, a two-day prototype. One test. A test that leads to a second test is deliberation in a lab coat.

The memo for the decisions that lock

A one-way door earns one page, written by the person proposing to walk through it, and read by everyone it binds before a decision date fixed in advance. The page has five parts and takes an evening.

The decision in one sentence, with the date by which it will be made and the name of the person who will make it. The alternatives, at least three, one of which is to do nothing, each with its own cost and its own reversibility. Writing them down is how you find the one you had not considered, which is often the smaller, reversible version: a contractor before a hire, a pilot before a contract, a month of working together before a co-founder agreement. What would have to be true for each option to be the right one, in terms you could check later. A premortem, which is Gary Klein’s technique from a 2007 piece in Harvard Business Review: assume it is a year from now and the decision has failed, and write down the most likely reasons why. People will say in a premortem what they will not say in a meeting, because the failure is already assumed and nobody is being disloyal by describing it. The reopening condition: the one signal that would bring the decision back to the table, named now so that it cannot be invented later by whoever regrets it.

Send the page to the people it binds. Give them until the decision date to disagree, in writing. Then decide on the date, not before and not after, and record it. A memo is not a slower way of deciding. Most founders who write one find that it ends the decision in a week, where the undocumented version had been circling for a quarter.

The small decisions take meetings and the large ones get made at eleven at night. Sort them first and the meetings disappear.

The four ways founders get this wrong

The small one treated as large. Three founders and a forty-minute discussion of which invoicing tool to use. Bezos’s warning in miniature, and the cure is a rule: anything reversible and under the cost threshold is owned by one person and decided by the end of the day. If two founders disagree about the tool, the one who will use it most decides.

The large one treated as small. The equity split agreed over a coffee because it felt awkward to write it down. The first hire made because the candidate was available that week. The lease signed because the office was nice. Each of these is a one-way door walked through with a two-way door’s process. The first of them is the one Paul Graham’s founders named as their biggest surprise in What Startups Are Really Like: a co-founder’s character is tested more severely in a startup than in almost any other situation, which is exactly why the equity decision deserves the memo and the coffee does not suffice. The lesson on [vesting and the co-founder cliff](/library/vesting-and-the-cofounder-cliff) is the equity one; [whether to raise at all](/library/should-you-raise-at-all) is the capital one.

Re-deciding. A decision made and then reopened every time the week is bad is not a decision. It is a mood. The reopening condition in the memo exists to stop this: if the named signal has not appeared, the matter is closed, and the energy goes into making the decision work rather than relitigating it.

Deciding by stamina. The co-founder who keeps the argument going until the other one gives up has not won a decision. Bezos’s remedy is to name the misalignment and escalate it, and in a company with no one to escalate to, the equivalent is to write it down, set a date and let the person whose domain it is decide, with the other committing. Disagree and commit works in both directions, and it only works if it is sometimes you who commits.

A log, and a quarterly reading of it

Keep a decision log: a single document, one line per decision, with the date, the box it was sorted into, who decided and the expected result. Fast decisions get a line; slow ones get a link to the memo. It takes two minutes a day. Then on the first working day of each quarter read the quarter’s lines with the co-founders and ask three questions. Which fast decisions turned out wrong, and were they really reversible? Which slow decisions would have been fine made fast? And which decision was not made at all, and what did the delay cost? Twelve quarters of this and a founding team has something most companies never acquire, which is a record of its own judgement and a way to improve it.

Nothing here is legal or financial advice; the decisions used as examples, on equity, leases and capital, each have their own lesson and deserve their own professional. The two Bezos letters are below and take twenty minutes together. Read the 2015 one first.

Sources

1. Jeff Bezos, 2015 Letter to Shareholders, Amazon.com, Inc., filed with the SEC April 2016 — Type 1 and Type 2 decisions; one-way and two-way doors.
2. Jeff Bezos, 2016 Letter to Shareholders, Amazon.com, Inc., filed with the SEC April 2017 — High-velocity decision making: seventy per cent of the information, disagree and commit, escalating misalignment.
3. Gary Klein, Performing a Project Premortem, Harvard Business Review, September 2007
4. Paul Graham, What Startups Are Really Like, October 2009 — On co-founders being tested more severely than in any other situation.