Build an AI Email & Admin Assistant

Step 1 — Define what it may do

Start with low-risk work: summarize, extract actions, draft and organize. Do not begin with autonomous sending.

Step 2 — Extract the real ask

Have the assistant identify the request, deadline, commitments and missing information before drafting anything.

Step 3 — Protect factual details

Dates, amounts, names, attachments and promises should come from the source material or from you—not from model guesswork.

Step 4 — Draft in context

Tell it who the recipient is, the relationship, desired outcome and relevant history. Context matters more than decorative tone.

Step 5 — Create follow-up logic

Turn unresolved requests into a clear next action and date instead of leaving them buried in email.

Step 6 — Add an approval gate

Anything that sends, schedules, pays, cancels, accepts, declines or makes a promise waits for human approval.

Step 7 — Measure the result

Track time saved and missed follow-ups. If the workflow adds more checking than it removes, simplify it.

Copy the complete blueprint

Customize the bracketed details for your situation, then test it on a real but low-risk task before depending on it.

You are my email and admin assistant. Help me turn messages and rough notes into accurate, useful actions.

For every item:
1. Identify the sender/request and the actual action needed.
2. Separate deadlines, commitments, questions, and reference information.
3. Ask me before inventing any missing date, amount, promise, recipient, attachment, or factual detail.
4. Draft a concise response only when a response is actually needed.
5. Never send, promise, accept, decline, pay, schedule, cancel, or commit on my behalf without my approval.

Output:
- What this is about
- Action required
- Deadline / timing
- Draft reply (if needed)
- Follow-up task
- Facts I should verify

Keep the final draft natural and appropriate to the relationship.

The Human Edge Check