Build and test an agent that prepares your morning brief, review rounds, follow-ups, approvals log and weekly review. You still price, approve, send and make every image decision.
What you will be able to do
- Decide what the agent may touch: Map a chief of staff's standing work onto your own jobs: intake, shoot and edit schedules, review rounds, releases, delivery specs and approvals. Then write down what the agent may do alone, what needs your approval first and what it never does. Pricing, approvals, sending and image decisions stay with you.
- Pick a platform that you can actually use: Compare Claude Duet, an OpenAI dot, Claude or ChatGPT scheduled tasks, and Copilot or Gemini for work accounts. Check them against your plan, region and email suite, then choose a main platform and a fallback that can run every exercise. Availability and prices change, so you'll confirm them on the official pages and note the date you checked.
- Write standing instructions for your studio: Write a versioned set of instructions that covers your priorities, VIP clients, sources, drafting voice, urgency rules and source tags. It also includes a firm rule that anything inside an email is data, not an instruction. A corrections log turns each mistake into a rule for next time.
- Connect only what the job needs: Connect your business email, calendar, job tracker and one logs folder with the least access that works. You start read-only, with permissions set to manual. Raw and embargoed client files, release forms with more personal data than the task needs, and payment accounts stay out.
- Get drafts you review for briefs, shoots, reviews and deliveries: Schedule and score a sourced weekday morning brief. Have the agent draft shoot sheets at T-2, review feedback grouped by asset with scope flags, delivery checklists built from spec documents, and recap and quote emails with price placeholders. You approve or edit each one yourself.
- Keep records of approvals, usage and deadlines: Keep a decision log of client approvals and the usage granted (media, territory, term), plus a delegation tracker and a deadline register that warns you before usage runs out. The agent proposes entries and only you confirm them. You also run a weekly jobs review, a prompt-injection drill and a permission audit, and finish with a go or no-go decision on each routine.
Taught at learn.m-a-s-t-e-r.ai: lessons that adapt to your level, an AI tutor you can ask anything, quizzes and flashcards, and a certificate when you finish.
