Claude Projects, ChatGPT Projects, and NotebookLM all let you upload a set of files and ask questions grounded in them. For one person and a manageable set of documents this is excellent, and if that describes your situation you should use one. The question is what happens when the documents belong to a team and the set keeps growing.
That question also has an everyday version, in group chat and email. What gets worked out in a chat thread belongs to the people who were in it. Someone who joins in March cannot read their way into a conversation from January, and nobody forwards a scroll of messages to a new hire. What goes into an email is addressed and filed, and can be handed on years later.
A project is the private thread of knowledge work. It is quick, it is genuinely good at what it does, and it belongs to whoever made it.
Where projects are strong
They are instant. Drop in twenty files and start asking. The model reads the files directly, so for a small set the answers are complete and the citations point at the right passages. Nothing to set up, nothing to maintain.
Where the ceiling is
- They are personal. A project belongs to the person who made it. Colleagues make their own, upload their own copies, and the organization's knowledge splinters into private notebooks that drift apart.
- They have a size. Every project has a limit on files and on how much text the model can hold at once. Below the limit the model reads everything; near it, the tool falls back to retrieving passages, and you are back to the document chat problem of answering a smaller question than you asked. Organizational libraries cross that line quickly.
- They do not build anything. Each answer is computed from the files in that conversation and then forgotten. There is no record of Marcus Chen that accumulates across files, no list of studies you can filter, no history of what was known last quarter.
- They do not run on their own. Nobody is checking the feasibility pipeline every Monday unless someone opens the project and asks.
- They are not administered. No roles, no audit trail, no way to give the new hire access to exactly the clinical knowledge base and nothing else.
What Desk adds
Desk is the shared, structured version of the same idea. Files are uploaded once for the whole team. Each is read into records, the links between them, and cited facts, so the knowledge base grows without a ceiling and questions that need everything, such as counts and comparisons, run over everything. Records persist and accumulate sources. Routines run scheduled checks and file the results. Admins control who sees which knowledge base, and every answer shows its sources.
A simple rule
If the files are yours, and there are fewer than a few dozen, use a project. If the files are the team's, keep arriving, and people ask "have we done this before" about them, that is what Desk is for. It is the same instinct that moves a decision out of a private thread and into an email somebody can be pointed to next year.
How Desk helps
Desk is a project that belongs to the team and does not stop growing.
- One upload for everyone, with roles and an audit trail, instead of private copies that drift apart.
- No file ceiling: each file becomes records and cited facts, so questions that need everything still run over everything.
- Records accumulate sources over time and routines keep asking on a schedule, so the knowledge base is more useful in month six than on day one.
Thinking about how this would work for your team?
Let’s chatKeep reading
"Have we done this before?" is the most expensive question in your company
It gets asked a dozen times a week, it is almost always answerable from files you already have, and the answer usually arrives too late to matter.
Read more →Tribal knowledge: what it costs and how to capture it without a reorg
The knowledge that walks out the door is rarely written nowhere. It is written somewhere unfindable. The fix is not asking people to document more.
Read more →