Give your assistant a small job, a clear brief, and boundaries. Make its work easy to review before expanding its responsibilities.
Write a small assistant brief
An AI assistant works better when you explain what matters, how you like information presented, and which decisions remain yours. Start with your working hours, typical commitments, preferred format, and the task you want help with.
Keep the brief practical and limited. You do not need to provide your complete personal history. For example: “I work best with three priorities, short checklists, and protected preparation time before meetings. Ask before moving a commitment.”
Give it a daily packet
Provide a short packet containing today’s fixed commitments, unfinished work, upcoming deadlines, and available time. Ask for a plan that distinguishes obligations from optional work. Tell it to explain any tradeoffs rather than silently deciding what matters most.
For a fictional day with two calls and one hour of focused work, a good assistant might suggest preparing the proposal outline and moving cosmetic website changes. It should ask if the website change fixes a serious problem before treating it as optional.
Use it before and after important moments
Before a meeting, ask for a brief with the purpose, known context, questions to resolve, and materials you should bring. Afterward, provide safe notes and ask for decisions, proposed actions, and missing details.
Before a difficult conversation, use it to rehearse a clear opening and anticipate reasonable questions. Afterward, write your own account of what happened before asking for interpretation. This helps keep the AI from replacing your memory with a persuasive story.
Separate advice from access and action
An assistant can help plan from information you paste without having access to your accounts. Calendar changes, email sending, file edits, and recurring tasks depend on actual tool capabilities, configured permissions, and appropriate authorization.
Before connecting anything, define what it may read and what requires approval. Begin with the smallest useful access. Check the confirmation after an action and periodically review the access you have granted.
A good boundary is explicit: “Draft responses for review; ask before sending, purchasing, deleting, or rescheduling.” The wording helps clarify intent, but actual permission controls and review still matter.
Keep a weekly review
Once a week, assess what the assistant helped you finish and what caused extra work. Remove routines you ignore. Update preferences when your circumstances change rather than expecting the system to infer them.
Watch for a subtle failure: a beautifully organized plan that treats every incoming request as equally important. The assistant should help you choose, including saying no or delaying work. You remain responsible for commitments and decisions.
Prompts to put it into practice
Replace the brackets with your context. Use information you have permission to share.
Create my assistant brief
Help me write a concise assistant brief from these preferences: [preferences]. Include priorities, working constraints, preferred output style, missing-information rules, and actions requiring approval. Ask questions rather than assuming access to my accounts.
Prepare my daily packet
Organize the commitments below into fixed appointments, deadlines, focused work, and open questions. Propose three priorities that fit [available time]. Explain tradeoffs and flag anything you cannot safely prioritize without more context.
Try it on your work
Use a short assistant brief for three days. Keep a note of one helpful result and one correction each day, then revise the brief.
Check the result
- The assistant has only the information and access it needs.
- Important actions have clear approval boundaries.
- My plan reflects my priorities rather than simply the newest request.
