Most people message an AI like a search box. John works better when you message him like a colleague who owns outcomes.

1. Lead with the outcome

Weak: "Can you look at my email?"

Strong: "By 9am, summarize overnight mail and draft replies for anything tagged urgent. Hold sends until I say OK."

John plans steps from the outcome, not from vague intent.

2. One goal per thread

Mixing "fix the Acme proposal" and "book dentist" in the same chat pollutes context. Start a fresh message for a fresh goal. John keeps memory on the server, but a clean thread keeps today's work legible.

3. Name your approval bar

Say it once: "External sends need my OK." or "Book under two hours without asking."

After that, John knows when to report done versus when to pause for you.

Bonus: close the loop

When John reports back, reply with a one-line correction or "ship it." That feedback trains the next task faster than re-explaining your preferences from scratch.

Try today: Send one outcome-first message before your next meeting block.

John reads a persona file on your private server before every task. You do not need a novel. You need clear defaults.

The five-line template

Copy this into your persona settings and edit in plain language:

• Role: Who John is for you (chief of staff, EA, ops lead, listing coordinator).
• Tone: Short and direct, warm with clients, formal with investors, etc.
• Priorities: What always comes first (revenue threads, hiring, active deals).
• Never do without asking: Send external email, book meetings, change CRM stage.
• Weekly rhythm: Morning brief time, Friday pipeline review, etc.

Example

Role: Executive assistant for a solo consultant. Tone: friendly, no jargon, max three short paragraphs per reply. Priorities: retainer clients before prospects. Never send without my OK. Weekly rhythm: brief at 8am weekdays.

Why it matters

Without a persona, every draft sounds like a template. With one, John stops asking how formal to be and starts shipping work that matches how you already operate.

Next step: Write five lines, save, then delegate one real thread and correct the draft once. John keeps the correction for next time.

Most knowledge workers lose the first hour of the day to email. Not because the messages are hard, but because triage is repetitive: sort, prioritize, draft, defer.

The pattern that works

1. Define outcomes, not steps. Tell your coworker what "done" looks like: "Summarize overnight mail and flag anything that needs a reply today."

2. Keep humans on the risky sends. Let the AI draft; you approve before anything leaves your account.

3. Reuse your tone. Store persona notes so drafts sound like you, not a template.

What to delegate first

• Overnight summaries with urgency tags
• Draft replies for routine threads
• Follow-up nudges for quiet contacts
• Calendar conflicts surfaced from mail

What to keep

• Term sheets, legal, and HR-sensitive threads until you trust the workflow
• Anything that moves money

John runs on your private server, connects to Gmail and Calendar, and reports back in Telegram. You delegate in plain language; he closes the loop inside your tools.

Next step: Connect your apps, write a short persona, and delegate one recurring morning brief.