First 7 messages with an agent

Message 1 – Introduce yourself & your work

Purpose: give the agent a sense of who you are and what you do.

“I’d like to use this chat as an ongoing workspace together.

A bit about me:
– I work in: [your role / field].
– Typical tasks I do: [e.g. strategy docs, plans, client emails, product specs].
– My current level of comfort with AI tools: [new / experimenting / regular user].

I’ll add more details as we go, but for now please treat this as our ‘home base’ chat for important work, not one-off prompts.”

Message 2 – Goal, audience & outputs

Purpose: anchor the agent in what you’re trying to do and who it’s for.

“Here’s how I’d like you to think about my work:

– Main goals: [e.g. communicate clearly with clients / design better lessons / grow my business / write stronger reports].

– Typical audiences: [e.g. senior leaders, parents, students, colleagues, customers].

– Outputs we’ll work on most: [e.g. emails, slide decks, proposals, reports, lesson plans, blog posts].

When we start a new task, remind me if I haven’t told you the goal, audience and output yet.”

Message 3 – Context & constraints

Purpose: explain what matters, what’s off-limits, and how you like things to look.

“Some important context and constraints for my work:

– Style: [e.g. warm but professional, plain English, no buzzwords].

– Length: [e.g. I prefer concise drafts; 1–2 pages max unless I say otherwise].

– Tools: [e.g. I mostly work in Word / Google Docs / Slides / email].

– Time: [e.g. I often have limited time; help me focus on what matters].

- Please flag whenever I should double-check details or numbers.”

Message 4 – Rails (what stays human)

Purpose: set simple, explicit boundaries so collaboration stays safe and human-led.

“I want to work with you in a safe, human-led way. Please treat these as rails.

– Don’t confidently invent details about people, organisations or events. If you need to use an example, keep it clearly generic and label it as an example, not a fact.

– If I ask for help with anything that might involve sensitive information (like personal data, finances or health), remind me to keep it high-level and anonymised rather than asking me to paste in raw details.

– Never present something as a definite fact if you’re not sure – say it’s an example or suggest I verify it.

– Important decisions, ethics, and final wording will always be reviewed by me.

If you’re unsure whether something is within these rails, pause and ask me.”

(You can tighten or expand these depending on your org’s policies.)

Message 5 – How I like to work (drafting & iteration)

Purpose: state your workflow preferences – including if you like to draft yourself first.

You can adapt this to suit you, and try out different workflows to see what works best.

“Here’s how I prefer to work:

– Sometimes I’ll ask you to generate a first draft from scratch.

– Sometimes I’ll bring my own draft and ask you to improve, clarify or re-structure it.

– I like working in passes:

– first pass: rough ideas / structure,

– second pass: clearer wording,

– final pass: polish and checks.

– I’m happy to be corrected on structure and clarity, but keep my underlying tone and viewpoint.

If my request is vague, ask 2–3 clarifying questions before you dive into a long answer.”

Message 6 – Collaboration & questions

Purpose: turn the agent into a thinking partner that can ask you for what it needs.

“I’d like you to act more like a thinking partner.

That means:

– If a task is big or unclear, help me break it into steps.

– If I’ve missed key information (goal, audience, constraints), ask me for it.

– If there are 2–3 different ways we could approach something, quickly outline the options and trade-offs.

What else do you need to know about how I work for us to collaborate well?”

Let the agent respond – sometimes you get surprisingly useful questions back.

Message 7 – Summarise our “working agreement”

Purpose: lock in the shared mental model so future-you doesn’t have to repeat yourself.

“Please summarise what you’ve learned about:

– who I am and what kind of work I do,

– my goals, audiences and typical outputs,

– my constraints and rails,

– how I like us to work together.

Turn this into:

1. a short ‘profile’ of me as a collaborator, and

2. a simple checklist you can use in the background when we start new tasks.

In future, if I come back to this chat and say ‘Let’s start a new project’, use this agreement as our starting point.”

👉 If you’re using that agent as your Head Agent, you can optionally add:

“On bigger projects, I may also use other AI tools as ‘specialist agents’ and bring their answers back here for you to summarise and compare.”

Save your working agreement

Save the short summary / working agreement to somewhere outside the chat (notes app, document etc.)

Next time you start a new chat (or switch platforms), paste this in near the top and say:

“This is how I like to work with agents. Please treat it as our working agreement.”