A new hire spends their first weeks asking questions that were answered, in detail, eighteen months earlier in a Slack thread that is now eighteen hundred messages deep in a channel nobody scrolls back through — a decision that was made, reversed, and made again by a team that no longer remembers the first attempt. The thread is technically searchable. It is not, in any practical sense, findable.
The common diagnosis is that companies have a knowledge problem, and Glean has built a real business on that diagnosis, raising large funding rounds on the bet that better enterprise search across Slack, Notion, and email will surface the right document at the right moment. The more accurate diagnosis, and the one this project is built around, is that most companies have a forgetting problem that search alone does not fix: information decays in accessibility almost as fast as it is generated, regardless of how good the index underneath it is.
Search treats every document as equally available at every moment, which is not how organizational memory actually degrades. A decision rationale is highly relevant the month it is made, moderately relevant for a year while people still remember the context around it, and then effectively invisible — not deleted, just no longer connected to anything a person would think to search for, and no search ranking fixes that on its own.
The job this tool actually has to do is narrower and harder than indexing text well: track which decisions are still active, which have been explicitly superseded, and which conversations resolved into a commitment versus which simply trailed off unfinished. A new hire does not need every mention of a topic ranked by relevance. They need to be told, plainly, which of three conflicting answers in the archive is the one still in force.
The people most affected by getting this wrong are not executives, who tend to have long institutional memory and direct access to whoever made a decision originally. It is everyone several layers removed from that original context — new hires, cross-functional collaborators, and anyone joining a project midstream — who inherit decisions with no visible reasoning attached.
The harder problem is not technical. It is convincing a team that the tool should surface a reversal from eighteen months ago even when nobody asked it to and the reversal is mildly embarrassing for whoever made the original call. A tool that only ever produces flattering, tidy answers has built a nicer search bar. One willing to say 'this was decided, then undone, and here is why' has actually built a memory.