Free resource
Give AI enough context to finish one thing without giving it permission to do everything.
Works with ChatGPT, Claude, Gemini, Messages, email, documents, and future agents.
Put it to work
Use this whenYou have one repeated task and one source that contains enough information for a reviewable draft.
Skip it whenThe task includes confidential data, an unclear recipient, a consequential decision, or an action you cannot reverse.
Choose one conversation, document, spreadsheet, folder, or approved system. Do not start with an entire inbox, drive, or account.
Ask for one draft, checklist, comparison, update, or finished file that you can review in a few minutes.
Say what the AI may prepare and what it may not send, publish, delete, change, purchase, schedule, or promise.
Tell the AI when missing information, an unclear recipient, conflicting instructions, or a new decision should end the task and bring it back to you.
If no existing step, report, or recurring task becomes simpler, shorter, or unnecessary, the AI may be adding another checkpoint instead of reducing work.
Decide what you need to see before the work counts as complete: source details, a draft, a changed-file list, a preview, or a confirmation screen.
Use only [SOURCE] to prepare [OUTPUT]. This handoff replaces [OLD STEP]. Do not send, publish, delete, change the source, make a purchase, or make commitments for me. Stop and ask me if the source is incomplete, the recipient is unclear, instructions conflict, or the task requires a decision I have not made. Before I approve anything, show me [PROOF OF COMPLETION] and separate what the source directly supports from anything you inferred.
One useful result exists, you can trace it back to the source, nothing consequential happened without your review, and one old step is smaller or gone.
The first handoff was accurate, saved real repetition, and stayed inside the boundary without extra supervision.
Three filled-in examples
Use only my Messages conversation with Sam to draft a reply to the latest question about Tuesday's arrival time. This replaces rereading the full thread and drafting from scratch. Do not send it or add a promise. Stop if more than one time appears. Show me the exact message you used and the draft.
Use only the approved launch plan to prepare a checklist of incomplete tasks. This replaces manually copying open tasks into a second list. Do not change the plan or assign work. Stop if a task has no owner or due date. Show me the plan section behind each checklist item.
Use only the current pricing record to draft corrections for the website. This replaces comparing every page from the beginning. Do not publish or change live pricing. Stop if the website and record conflict. Show me the current text, proposed text, and source line before any edit.
Read the original issue
ChatGPT reached Apple Messages while Anthropic widened its agent-building layer. The Brief explains what changed, what can help, and why the final check still belongs with you.