The annoying part
when i work on a project, i don’t work in one perfect conversation.
i might be in Claude chat thinking through the shape of the thing. then Claude Code is changing files. then Claude Cowork is helping with a follow-up. then maybe an occasional ChatGPT or Codex thread wanders in because the tool for the moment is different.
and every time i switch rooms, i end up carting around the same project details: what we decided, what broke, what names mean, what not to touch, what the next step was, why something is weird on purpose.
that is the part that gets old. not because Claude is bad at context, but because the context is scattered across twenty conversations that do not share a clean project memory.
Teams make this worse
working with other people makes the same problem multiply — so team workspaces are now the heart of agent-wiki.
everyone has their own Claude chats, their own Claude Cowork sessions, their own Claude Code runs, their own little trail of “wait, where did we decide that?”
now you have 72 different conversations about one project, and each one has about one-fifth of the project knowledge. nobody is doing anything wrong. the knowledge just has no shared home.
agent-wiki is that home: a workspace for your team with a plain markdown wiki per project, which people and every teammate’s Claude can read, update, and leave behind for the next session. invite the team with a link, connect Claude once each, and let the maintenance agent keep the garden tidy.
Won’t Anthropic just build this and put you out of business?
i hope so!
that is part of why everything here is exportable. if Anthropic, or anyone else, makes this easier in the future, you should be able to take your project knowledge with you and cut ties with me.
that would be a win for everyone. including me, spiritually. my bank account may have notes.
the tl;dr is simple: this is a real problem i solved for myself, and i want everyone else to have it too.