Great piece. One question though, why not Claude Teams Projects for the shared context problem? You can create shared projects with custom instructions, upload reference docs, and give the whole team access to the same context. No symlinks, no Google Drive desktop installs, no "make available offline." It's not a local filesystem agent, but for the 80% of knowledge workers who aren't going to learn terminal commands, it seems like the simpler answer. Curious what you think is missing from it.
Hey Federico, great question. Claude Teams projects could also give a lot of the value, for sure. For a lot of teams it could be a very wise compromise, or starting point. None of the companies I've worked with chose that option (some were using Claude Projects and wanted to move off of it) so I don't have as much depth on it.
The best way to use AI at work today is working with a full-fledged file system (local or virtual), for many reasons: far faster learning loop, persistence, customizability, and individuals being able to keep their own private area for collaborating with AI.
That said, I'd trade all of that for adoption and having people truly see their work differently, which is most important. Long term, I hope the jankiness of a local filesystem goes away quickly so it becomes an easy choice.
(Btw, to use Claude Desktop doesn't require any terminal commands, they've done a great job with Cowork and even Claude Code as a GUI desktop app.)
Great points. We’re currently running a team experiment at my company, and we chose GitHub as our knowledge base. I like the “driving stick” analogy - that’s exactly how it feels for a PM who doesn’t code much.
However, since we have some highly technical people involved, we’ve moved beyond simply sharing knowledge or producing “knowledge artifacts.” We’re actually building things together (skills, MCPs, connectors, etc.), and GitHub has been very helpful at this stage.
Great piece. One question though, why not Claude Teams Projects for the shared context problem? You can create shared projects with custom instructions, upload reference docs, and give the whole team access to the same context. No symlinks, no Google Drive desktop installs, no "make available offline." It's not a local filesystem agent, but for the 80% of knowledge workers who aren't going to learn terminal commands, it seems like the simpler answer. Curious what you think is missing from it.
Hey Federico, great question. Claude Teams projects could also give a lot of the value, for sure. For a lot of teams it could be a very wise compromise, or starting point. None of the companies I've worked with chose that option (some were using Claude Projects and wanted to move off of it) so I don't have as much depth on it.
The best way to use AI at work today is working with a full-fledged file system (local or virtual), for many reasons: far faster learning loop, persistence, customizability, and individuals being able to keep their own private area for collaborating with AI.
That said, I'd trade all of that for adoption and having people truly see their work differently, which is most important. Long term, I hope the jankiness of a local filesystem goes away quickly so it becomes an easy choice.
(Btw, to use Claude Desktop doesn't require any terminal commands, they've done a great job with Cowork and even Claude Code as a GUI desktop app.)
Great points. We’re currently running a team experiment at my company, and we chose GitHub as our knowledge base. I like the “driving stick” analogy - that’s exactly how it feels for a PM who doesn’t code much.
However, since we have some highly technical people involved, we’ve moved beyond simply sharing knowledge or producing “knowledge artifacts.” We’re actually building things together (skills, MCPs, connectors, etc.), and GitHub has been very helpful at this stage.