Your team's memory resets every Monday.

Every week you re-explain the same decisions, re-find the same context, and re-discover what someone already knew. That's not a culture problem. It's a tooling problem.

Pick any Monday. Someone on your team asks a question in Slack that was answered three months ago. A new hire opens a pull request that redoes a decision already relitigated twice. A product manager writes a spec section that duplicates something in a Google Doc from last quarter — which nobody can find because nobody can remember what it was called.

This isn't a culture problem. Your team isn't bad at communicating. The tools just forget faster than you do.

The three ways teams lose memory

Docs rot. A wiki is a garden, and almost no company has a gardener. Pages that were true six months ago now subtly mislead the people who find them. The ones that are still accurate are hard to distinguish from the ones that aren't.

Slack buries. The conversation where you actually figured something out — the one with the dozen-message thread and the moment of clarity at message #9 — is seven scrolls back in a channel you joined three weeks ago. You can search, but only if you remember the right word.

Meetings evaporate. The hour where four people aligned on strategy leaves behind, at best, a bullet-pointed Notion page. At worst, it leaves nothing. The alignment lives in four heads and immediately starts drifting.

Every one of these is a failure of the substrate, not the people. Your team knew the answer. The tools couldn't hold it.

What "team memory" actually means

A real team memory layer has three properties that docs, Slack, and meetings don't.

It captures without asking. If your memory system requires a human to remember to write things down, you've just moved the problem. It has to ingest from where work already happens — Slack, meetings, docs — and build the memory passively.

It surfaces at the right time. Storing is easy. The hard part is knowing when to say "hey, your team already talked about this — here's the decision and the reasoning." That's not a search problem. It's a relevance problem, and it requires a model of who's asking what.

It's queryable by humans and agents. If your AI agent opens a ticket without knowing the decisions that shaped the codebase, it writes the wrong fix. Team memory is as much infrastructure for agents as for people. Both need the same substrate.

Why this gets more urgent, not less, with LLMs

You might think LLMs make this problem better. More automation, less re-explaining. In practice, the opposite is happening: every agent you spin up is an amnesiac. It needs context to do anything useful, and the more agents you have, the more surfaces you have to brief. Without a shared memory layer, you end up re-explaining your codebase, your product decisions, and your team's idioms to every new agent conversation, forever.

The companies that win the next decade aren't the ones with the best prompts. They're the ones with the best memory — the ones whose humans and agents inherit context instead of re-earning it every Monday.

That's what we're building with Base. Team memory that compounds instead of resets.