A memory that keeps everything can lead you to worse decisions than no memory at all
I have been building Verbasil, a company memory, for two months. And the most counter-intuitive lesson of these two months does not come from what the memory forgets. It comes from what it brings back.
Seven dead things, presented as alive
In early August, using Verbasil on my own company, I ran into this: seven items to follow up on were in fact already settled. But nothing had ever closed them. So they sat there, and Verbasil's chat kept citing them as ongoing topics.
What stopped me was not the mistake. It was the tone. The machine had no doubt. It presented dead things with the same confidence as living ones, because inside its memory there was no difference: everything was there, so everything was true.
Being confidently wrong costs more than forgetting
An empty memory has a virtue we underestimate: it stays silent. When you do not know, you check, you ask again, you go back to reality. That is why the title of this article is not an exaggeration: an empty memory forces you to verify, a stale memory excuses you from it. The first costs you time. The second costs you decisions.
Because a memory that keeps everything always has an answer. And when a decision has changed, when a choice has been reversed, it has two answers: the old one and the new one, side by side, with equal weight. Nothing tells the machine which one is alive.
That is the real trap, and nobody warned me about it: not forgetting, but false confidence. A partner who tells you "I am not sure anymore" is doing you a favor. A partner who repeats a cancelled instruction, in a tone of complete certainty, leads you into bad decisions with total peace of mind.
Three engineering choices that came out of it
These situations, met while building and using my own product, produced three design rules. They are real choices in the engine, not decorative principles.
Undoing is a first-class memory operation. Archiving is not deleting. When a source is archived in Verbasil, everything derived from it expires with it, stamped with the date of that action. And if you undo it, only what that precise action switched off comes back on. This detail sounds technical; it decides everything. Without it, unarchiving would resurrect decisions that were put away by hand weeks earlier, and the memory would become a liar through another door.
What sleeps should speak up, instead of waiting. An item that has not moved in a month is no longer information, it is a question. So the product asks it: done, keep it, or no longer relevant? Three outcomes, one gesture. That is the direct answer to the seven dead items above: they were not waiting for better search, they were waiting for someone to ask whether they were still true.
When in doubt, warn rather than assert. When the source behind a decision has been rewritten since, no machine can automatically tell whether the decision still holds. We chose not to guess: the decision carries a flag, "source edited, re-check", and the human decides. An honest memory knows how to say "worth verifying".
The test I suggest
Look at your own company's memory, whatever shape it takes: your wiki, your notes, your team channel. And ask yourself one question: what is still presented there as alive when it is actually dead?
If the answer is "I do not know", that is normal. Nobody does this sorting, because no tool ever asks for it. We pile things up, and we trust the pile.
That is the conviction these two months gave me: the job of a memory is not to keep. It is to know what is still true, and to be honest about the rest.
If the underlying topic interests you, I have written about why startups forget their decisions and about how to keep track of decisions without turning it into a chore.