Product decisions that don't evaporate between Mondays

The decision was made. Someone said it clearly, in the right meeting, with the right people in the room. By Thursday it's effectively gone — not because anyone forgot, but because it was never written somewhere that any AI tool or future session would actually load.

This is the texture of most product work. Decisions get made in Slack threads, in standup comments, in the margins of Figma files. They're real. They shape what gets built. And they exist nowhere that your AI tools can access when you sit down to write the next PRD or plan the next sprint.

Where decisions go to die

The standard capture path for product decisions is: meeting → someone's notes → Notion page that nobody opens → irrelevance.

The problem isn't discipline. Most product teams are diligent. The problem is that "written down somewhere" and "available in every future AI context that touches this topic" are entirely different things. A Notion page that's never linked to a PRD session, never surfaced in a roadmap review, never loaded in a sprint planning discussion — that page might as well not exist from the perspective of your AI tools.

Every Monday standup where someone asks "wait, didn't we decide this?" is evidence of the gap. Every PRD that re-litigates a choice made three sprints ago is evidence of the gap. Every new PM who spends their first month absorbing institutional knowledge that could have loaded on day one is evidence of the gap.

What changes when decisions live in memory

When a product decision is captured as a Memory Stack memory — tagged to the relevant project, feature area, or initiative — it becomes part of the context that loads whenever someone works in that area.

The next PRD session for the affected feature starts with: here's the architectural constraint we agreed to, here's the scope call we made last sprint, here's the performance tradeoff we accepted in Q1. Not as background reading to do before the session. As loaded context at the start of the session.

This changes the quality of what gets produced. A PRD written with the last six months of relevant decisions available in context is more coherent than one written from scratch. A roadmap review that can surface "we already decided not to pursue this direction, here's why" in real time moves faster than one that depends on someone's memory.

Stale decisions surface automatically

One of the quieter features is what happens to decisions over time. A constraint agreed to in Q1 may no longer apply in Q3. A scope call made in January may need revisiting given what you've learned since.

The Stale Shelf system surfaces memories that haven't been accessed in a configurable period — flagging them for review rather than letting them quietly continue governing decisions they may no longer be relevant to. The team member who owns the decision sees it surfaced, reviews it, and either confirms it's still operative or updates it.

This is different from decisions silently expiring in a Notion page. The memory stays live until someone explicitly retires it — and the system makes the retirement prompt visible rather than depending on someone to go looking.

The PM turnover case

The most expensive version of the decision-evaporation problem is PM turnover. When a PM leaves, the institutional context they carry — the backstory behind every scope decision, every technical tradeoff accepted, every feature that was explicitly descoped and why — goes with them.

A successor PM starting with rich product memory starts differently. The decisions don't need to be reconstructed from ticket history and Loom recordings. They're in the memory layer, attributed, dated, and available on day one. The ramp-up shortens. The re-litigation shortens. The risk of reversing a deliberate decision because nobody knew it was deliberate goes down.

What good capture looks like

Not every standup comment needs to be a memory. The signal worth capturing:

• Scope calls with rationale: "We're not building X in this cycle because Y. Revisit after Z."
• Tradeoff decisions: "We chose approach A over B. The reason was C. The cost was D."
• Explicit deferrals: "This is a real problem. We're not solving it now. Here's why."
• Constraint acknowledgements: "This decision is constrained by the platform limitation in area X."

The common thread: decisions where the reasoning matters, not just the outcome. The outcome is often in the ticket. The reasoning is what evaporates.

The honest version

Memory is as good as what's been captured. A team that makes good decisions but never logs them will have sparse context. The passive capture system helps — it identifies decision-relevant content as it flows through AI sessions and surfaces capture candidates — but it doesn't catch everything, and it doesn't retroactively surface decisions made before Memory Stack was in the stack.

The compounding value is forward-looking. Start capturing now; every session that follows starts with more context. The institutional knowledge that builds in the first six months makes the sessions in months seven through twelve qualitatively different.

Product work runs on accumulated context. The decisions made in Monday's standup should still be available in Thursday's PRD session, in next month's roadmap review, and on the day a new PM joins. The infrastructure for that has existed. Most teams just haven't connected it yet.