The tax you don't notice
Every time you start a session by re-explaining your architecture, conventions, and constraints, you pay a small tax in time and mental focus. Because it's spread across countless sessions, it never shows up as one big cost — but it adds up.
It's the classic context-switching cost, applied to working with a stateless agent.
It costs quality too
Re-explaining from memory is error-prone. You forget to mention a constraint, summarize a decision loosely, and the agent acts on an incomplete picture — producing work you then have to correct.
So the cost isn't only the explaining; it's the mistakes that follow incomplete explanations.
Why it stays hidden
A two-minute re-explanation feels trivial in the moment, so nobody optimizes it. Multiply it by every session, every teammate, and every agent, and it's a meaningful drag on velocity.
Hidden costs are the ones worth measuring precisely because they hide.
Pay it once with persistent memory
A maintained project memory file pays this cost a single time: you encode the context once, and every session (and every agent) starts informed without re-explanation.
It converts a recurring tax into a one-time, reusable investment.
The compounding upside
Remove the re-explanation and every session starts faster and more accurately. Across a team and many sessions, that compounds into real velocity — and fewer corrections.
It's one of the highest-leverage small habits in AI-assisted development.