Review & approve

The Associate can act on your project — but it never changes anything on its own. It works by proposing, and nothing takes effect until you approve it. This is the heart of how Bluprint keeps an AI that can act from becoming an AI you can't trust.

How it works

When the Associate wants to make a change — create tasks, adjust a timeline, update statuses — it doesn't write straight to your project. It stages the change: a proposed set of edits, with a plain-language summary of what it intends to do and why.

You then decide:

Accept — apply the staged changes to your project.
Decline — discard them; nothing happens.
Remove individual changes — keep the parts you want, drop the rest, then accept the remainder.

Until you accept, your project is untouched. You're always the one who commits.

Not a black box

Because you see the proposed change before it happens — the what and the why — you're approving a concrete action, not trusting a mystery. That's the difference between supervised action and blind automation.

Built-in safety nets

The approval step is backed by a few guarantees:

Proposals expire. If you don't act on staged changes, they expire on their own (24 hours by default) rather than lingering — so nothing gets applied long after the conversation that created it.
Permissions are re-checked at the moment you approve. Even if you could make a change when the Associate proposed it, Bluprint verifies you still have permission to write to that project when you accept. If your access changed in between, the change is refused. It fails closed, never open.
Everything is scoped to your company. Staged changes are isolated per company and per project — they never cross tenant boundaries.

Advisory-only mode

If you'd rather the Associate only advise and never offer to make changes, turn tool execution off with /tools off. It becomes a pure thinking partner — analysis and answers, no actions. Turn it back on with /tools on.

Your call, every time

Whether it's one task or a whole sprint plan, the Associate's job ends at the proposal. Applying it is always your decision.

Last updated: 2026-05-31

PreviousHow it works Next Permissions & data

How it works

You then decide:

Accept — apply the staged changes to your project.

Decline — discard them; nothing happens.

Remove individual changes — keep the parts you want, drop the rest, then accept the remainder.

Until you accept, your project is untouched. You're always the one who commits.

Not a black box

Built-in safety nets

The approval step is backed by a few guarantees:

Proposals expire. If you don't act on staged changes, they expire on their own (24 hours by default) rather than lingering — so nothing gets applied long after the conversation that created it.

Permissions are re-checked at the moment you approve. Even if you could make a change when the Associate proposed it, Bluprint verifies you still have permission to write to that project when you accept. If your access changed in between, the change is refused. It fails closed, never open.

Everything is scoped to your company. Staged changes are isolated per company and per project — they never cross tenant boundaries.

Advisory-only mode

Your call, every time

Whether it's one task or a whole sprint plan, the Associate's job ends at the proposal. Applying it is always your decision.