Claude Artifacts sharing limits, and what to use when a team needs more

A Claude Artifact has two sharing states as we have used it: only me, or anyone with the link can edit. There is nothing in between. There is no member list, no role, no permission that says “this person may comment but not change anything”, and no way to give an AI agent an identity of its own that you can later take away. That is fine for a document or a prototype. It becomes a problem when the Artifact has quietly turned into your team’s backlog.

This piece describes that gap fairly, explains when it matters and when it does not, and says what to use when it does. We should be open about our interest: fenbs exists because of this gap. The first test we set ourselves was to move our own backlog off an Artifact and onto a fenbs board, and to stop if the board was not better.

What Artifacts are good at

Artifacts are an excellent way to make something with Claude and look at it straight away: a small app, a chart, a document, a tracker. A lot of people, ourselves included, have ended up with a backlog living in one, because asking Claude to “make me a board for these tasks” works, and the result is pleasant to use. For one person, it can stay that way for a long time.

Nothing here is a criticism of Artifacts. They were built for making and sharing things, not for running access control over a team’s work. Anthropic may well change the sharing options, so check Claude’s own help pages for the current state before you rely on any description of it, including this one.

Where the two states run out

Only me

While the Artifact is private, the backlog is yours. Your colleague cannot add the bug they found. Your client cannot look at progress. Everything comes through you, by message or email, and you type it in. That is the “can you write that down somewhere” problem, and it is how things get lost.

Anyone with the link can edit

Once you share it for editing, the link is the permission. Whoever holds it can change anything, and a link is easy to forward. You cannot give the client a smaller set of rights than the developer. You cannot let a tester report a bug without also letting them rearrange everything. You cannot remove one person, because there are no people, only the link.

What is missing in both

• Members. A list of who is on the board, so that adding and removing a person is a thing you can do.
• Roles. A way to say this person can read and comment, that one can add tasks, this one can change everything.
• Attribution. A record of who changed what, which needs named members before it can exist at all.
• Identities for AI assistants. If an assistant edits the shared thing, there is no separate key for it that you can scope or revoke.

When it matters, and when it does not

If you work alone and nobody else needs to see or touch the list, you may not need anything else. A private list is a private list.

It starts to matter when any of these become true:

• Somebody other than you finds the bugs, and you want them to file them directly.
• A client wants to look at progress whenever they like, and should not be able to change anything.
• More than one AI assistant works on the list, and you want to know which one did what.
• You need to take one person’s or one assistant’s access away without affecting everybody else.
• Someone asks “who moved this to done?” and nobody can answer.

What the missing layer looks like

fenbs is that missing layer, built as a product. It is deliberately small. There are four words to learn: board, task, lane and member. A task is a Feature, an Enhancement or a Bug. A lane is To Do, Next Up, In Progress or Completed. A member is a person, or an AI assistant working on a person’s behalf, holding a role on the board.

People get roles

You add your team, your client, or a friend who is testing for you, and each gets a role. You can give them a smaller one than yours. The fenbs home page describes five: an Owner and an Admin who run the board, an Editor who can create, edit, move and comment on any task, a Reporter who can create tasks and comment but edit only their own, and a Viewer who can read and comment. Take someone off and they stop seeing the board, while what they wrote stays where it is.

Assistants get keys cut from yours

An AI assistant connects with a token you issue under Settings. It acts as you and inherits your role, narrowed further by scopes you tick: read, write and comment. It is shown once, it can never do more than you can, and revoking it takes one click without signing you out. The steps are in the connection guide.

Every change is signed

Every change is recorded with who made it, person or assistant, and each task shows its own trail. This is the part a shared document cannot give you, because a shared document does not know who its editors are.

Moving a backlog across

Nobody adopts a tool by retyping a list they already have. On a fenbs board, Add many sits next to + New task. Paste one task per line, or paste a markdown list, which is what a backlog in an Artifact usually looks like when you copy it out. Start a line with bug:, feature: or enhancement: to set that line’s kind. Indented lines under an entry become that task’s note.

## Checkout
- [ ] bug: Search ignores the category filter
      Only at 390px. The chip renders but the query never carries it.
- [ ] Add keyboard shortcuts
      j and k to move between cards.

Everything is previewed before anything is created, so a line that was read wrongly is something you correct rather than clean up afterwards. Each task then gets a ref such as BU-016 that is never reused, even after a delete, so a ref quoted in a commit message or a text to a client keeps meaning the same task.

A fair summary

Keep using Artifacts for what they are good at: making things with Claude and showing them to people. When the thing you made is a list of work that other people and other assistants need to touch, with different rights, and you need to know afterwards who did what, you have outgrown a link. That is the point to move the list somewhere with members and roles. How it works shows the whole of fenbs in five steps, and plans start at $4.99 a month on the pricing page.