Reconstruction Tax
05-sep-2026
I have noticed something while building with AI.
We are getting incredibly good at moving fast. A problem comes up, we discuss it, make a decision, build something, test it and move on. At that moment everything is clear because the entire journey is still in our head.
Then a few weeks pass.
You come back to the same piece of work and something strange happens. The files are still there. The code is there. Sometimes even the decision is written down. But you no longer remember why you made that decision in the first place.
So you start digging.
Old chats. Notes. Git history. Screenshots. Maybe an email. You slowly reconstruct what happened until the decision begins to make sense again.
I realised that I was paying for the same understanding twice.
The first time when I learned it.
The second time when I had to reconstruct it.
I started thinking of this as Reconstruction Tax.
What I mean by Reconstruction Tax
Reconstruction Tax is the cost of having to rediscover something that we once understood because we did not preserve enough of the context around it.
The important part is that nothing necessarily has to be lost.
The code may still work. The document may still exist. The decision may still be sitting in a repository.
What disappeared was the chain around it.
What were we trying to solve? What did we know at the time? What did we assume? What did we test? What failed? Why did we choose this path and not another one?
Without that chain, the final answer survives but some of the knowledge behind it does not.
And I am beginning to think that this happens everywhere.
A person leaves a company and suddenly a process nobody questioned becomes difficult to understand. A founder looks at a decision made six months earlier and remembers the conclusion but not the reasoning. An AI helps us build something very quickly, but the reasoning remains buried inside a conversation that nobody will realistically read from beginning to end again.
Each time, somebody pays the Reconstruction Tax.
Why Ravi Wiki made this clearer to me
A large part of building Ravi Wiki was actually about this problem without me initially calling it Reconstruction Tax.
We kept asking simple questions.
If this computer disappeared tomorrow, what would remain?
Could another clean machine reproduce what we had built?
If an experiment worked once, did we preserve enough evidence to understand why we trusted it?
If a future AI joined the project with none of today's conversation history, could it reconstruct the important decisions from the repository?
That slowly changed my understanding of what it means to possess knowledge.
Knowing something once is not the same as possessing it durably.
For important things, some part of the reasoning has to survive us forgetting it.
That is also one of the reasons this wiki exists.
I don't want Ravi Wiki to become a place where I dump everything I read or think. That would create another problem. I want it to preserve the ideas that I keep returning to, where they came from, how my understanding changed and what they connect to.
In that sense, the wiki itself is an attempt to lower my own Reconstruction Tax.
The problem on the other side
There is an obvious danger here.
The answer cannot be to document everything.
If every decision needs five pages of explanation, documentation itself becomes work and eventually nobody wants to maintain it.
So there are probably two taxes.
Too little preservation creates Reconstruction Tax.
Too much preservation creates Documentation Tax.
Documentation Tax is the cost of preserving so much that maintaining the record becomes its own work.
Somewhere between them is the useful amount of memory.
I don't know exactly where that point is yet.
A question I want to keep exploring is:
What is the smallest amount of context worth preserving today so that tomorrow's reconstruction is cheap?
Maybe that is the real discipline.
Not remembering everything.
Remembering enough that we do not have to start again.
Related ideas
-
Documentation Tax — the opposite failure: too much preservation.
-
Builder's Journal — preserving the experience and meaning around work before hindsight smooths it away.