The First Version Is Always a Confession
Why the plan is always more confident than you are
A blueprint is a confident document. Every line is deliberate, every dimension resolved. It carries no trace of doubt, because doubt does not survive the act of drawing — you cannot render an ambiguity at 1:100 scale. By the time an idea reaches the page, it has been quietly stripped of everything its author was unsure about.
This is not a flaw. It is the function. A plan is a theory of what will matter, asserted with more certainty than its author actually possesses.
My PRD for Family Pay was a confident document. The prototype had other ideas.
The gap you cannot reason your way to
Institutions run on two layers — a formal one they can measure and an informal one they cannot. The consequential intelligence tends to live in the second. A credit model encodes what the institution chose to write down; the underwriter’s judgment encodes everything it didn’t.
The uncomfortable part is what this implies about knowledge itself. If the important intelligence is the part that resists formalization, then no amount of thinking — however rigorous — can fully surface it in advance. You can reason about the gap. You cannot reason your way across it.
Building is the only instrument I know of that measures the gap directly.
When you ship something, you are not releasing a finished idea into the world. You are releasing a hypothesis about what mattered — a formalization of your best current theory — and watching reality mark it up. The corrections come back not as arguments but as behavior. People use the thing for something you did not design. They route around the feature you were proudest of. They break it in a place you were certain was load-bearing and ignore the place you reinforced.
Every one of those is information you did not have, and could not have had, until the artifact existed to provoke it.
Why the first version specifically
There is a reason the first version is the one that confesses.
A mature product has already absorbed its corrections. Its scars have closed over; the workarounds have been paved into features; the original theory has been quietly revised so many times that you can no longer see where it was wrong. The gap is still there, but it has gone informal again — it has migrated back into the tacit knowledge of the team that maintains the thing.
The first version has none of that cover. It is your theory of what matters, exposed at full resolution, before reality has had a chance to edit it. The distance between what you built and what was actually needed is never as legible as it is in that first contact. After that, the system starts hiding its own mistakes from you, the way institutions do.
This is why building teaches something that thinking cannot. Not because thinking is inferior — but because thinking has no mechanism for being surprised. A plan can only contain the questions its author already knew to ask. An artifact, released, generates the questions no one knew were there.
The corrections Family Pay sent back
Family Pay was a product I designed from scratch — a UPI-first app for Indian teenagers, with category-based spend controls, goal-based savings, and parental oversight. The PRD was detailed. I had a clear theory of how each piece would work.
The prototype went to a few friends first. The corrections started before they finished testing.
The savings feature made the first one. My design had teens setting aside money toward specific goals — a pair of shoes, a school trip — with those amounts tracked within their primary UPI balance. What I had not considered: when you display a teenager’s full balance, including the portion they have committed to a goal, you are showing them spendable money. The mental accounting that separates saved from available does not survive contact with a number on a screen. The feature designed to encourage saving was, in its first form, an invitation to spend. The fix — a wallet-based savings bucket, segregated entirely from the spending balance — was a significant departure from the original design. It came not from a planning session but from building the interface and seeing what the balance display actually communicated.
The spend controls made a different confession. My PRD specified keyword-based filtering to block transactions at merchants selling prohibited items — alcohol, tobacco. This is a clean formal rule. Reality is not. A keyword filter that catches cigarette purchases also flags the corner chai shop, the kirana that sells chewing gum alongside everything else. The category prohibited merchant does not map neatly onto merchant name. A rule that looked rigorous on paper dissolved into three separate enforcement layers — MCC codes handled by the bank partner, keyword logic limited to specific contexts, ML for the rest — none of which had been fully visible when I was writing the specification.
Two features. Two corrections. Neither was available to the plan. Both arrived only once the artifact existed to provoke them.
The discipline this asks for
If the first version is a confession, then the instinct to delay it — to refine the theory further before exposing it — is a way of avoiding the only feedback that actually informs you. We tell ourselves we are not ready. More often we are simply unwilling to learn how wrong the plan was.
I held Family Pay in planning longer than I needed to. The word I used was refining.
This does not argue for carelessness. A confession still has to be worth making; you have to have built enough of a real theory to be meaningfully corrected. But it does reframe what shipping is for. The point of the first version is not to be right. It is to find out — precisely, behaviorally, unavoidably — where your formalization stopped matching the world.
Institutions lose their informal intelligence when they mistake a faster formal layer for a smarter one. The builder faces the inverse temptation: to mistake a more elaborate plan for a more correct one. Both are the same error in different clothing. Both confuse the legible part of the problem for the whole of it.
The plan is what you know. The first version is how you find out what you didn’t.