Choose what belongs in an AI task brief
A simple way to select relevant context for an AI task without sending every document your team owns.
When an AI assistant misses something, the tempting fix is to add more documents. Sometimes that helps. Sometimes it buries the instruction that mattered beneath unrelated material.
Choose context around the decision or deliverable the assistant needs to produce.
Build a small core
Start with the outcome, the intended audience, the current state, the constraints and the acceptance criteria. Include the authoritative sources needed to establish facts. A task to revise an invitation email needs the invitation behaviour and voice guidance; it probably does not need the entire product roadmap.
Separate instructions from reference material. A customer quote in a research document should not become a command to the assistant. Label examples, historical notes and current requirements so their roles are clear.
Add depth when the task needs it
Link supporting sources by subject and say when they matter. For example, ask the assistant to consult the permission model before proposing changes to who can invite a teammate.
This only works if the assistant can actually access those links. Check access at the start and ask it to identify missing sources before drafting a final recommendation.
Check understanding before execution
For a substantial task, request a short restatement of the goal, constraints and unresolved questions. Correct misunderstandings there rather than after a long implementation.
When work fails, identify the missing fact or ignored instruction. Add that specific information to the brief or improve its placement. A good brief grows through evidence about what the task requires, rather than accumulating every document that might conceivably be related.