TOOLKIT
Working with a small agent team
Type: Template
Purpose
This toolkit shows you a simple way to use more than one AI tool (ChatGPT, Claude, Gemini, Perplexity, Copilot, Grok, etc.) on the same piece of work – without it turning into chaos.
You don’t need any special setup or integrations, just a browser and access to the agents you already use (most of them have have a free tier you can start with.)
The basic pattern is:
You are the coordinator / editor
One Head Agent is your main project hub
2–3 specialist agents act as researchers, critics, or idea-generators
When to use this
Use a small agent team when:
The task is complex or important – a new service, strategy, policy, workshop, talk, report, etc.
You want different angles (research, critique, tone, style).
You don’t fully trust one model’s answer (which is sensible).
For simpler tasks, a single agent is fine. For bigger work, a small team gives you more depth and fewer blind spots.
A few rails (safety notes)
As with all AI work, keep some basic guardrails in place:
No confidential or sensitive data unless you’re sure about settings, policies and contracts.
Judgement, values, ethics and final decisions stay human.
Assume each model has its own biases and blind spots – that’s partly why you’re using more than one.
The roles
✍️ You – Coordinator / Editor
You act more like a Chief Editor than a “user of tools”:
Set the goal, audience, constraints.
Decide which agent to use for what.
Choose what to keep, adapt, or discard.
Make sure the final output fits your standards, context and voice.
Head Agent – Project Hub
Pick one agent to act as your Head Agent – usually the one you’re most comfortable with (for us, that’s Sol / ChatGPT).
Their job is to:
Hold the overall context of the work.
Help you write good prompts for the other agents.
Read and summarise their answers.
Suggest options, trade-offs and next steps.
Specialist Agents – Guest Experts
Treat other agents like guest experts you invite into a meeting.
Examples (you can adjust for your own setup):
Research:
Gemini – broad web research, Google ecosystem, good with images and videos.
Perplexity – deep web research with citations and multiple sources.
Strategy:
ChatGPT, Claude – framing, structure, strategy, tone, stress-testing ideas.
Culture signals:
Grok – what’s happening on X, cultural context and sentiment
You don’t need all of them, pick 1–3 that match the task. The same agent can also wear different “hats” (researcher, critic, stylist etc.) depending on what you ask it to do.
A simple workflow
-
1. Brief your Head Agent
Pick your Head Agent and make the role explicit.
“You’re my main agent on this project. I’ll also be using other AI agents.
Your job is to:
– help me write prompts for them,
– read their responses,
– and summarise the best bits back to me in plain language.”Then give context:
Goal: What are we trying to do?
Audience: Who is this for?
Constraints: Deadlines, tone, length, non-negotiables.
Other agents: Which ones you’ll be using and what they’re good at.
👉 This helps the Head Agent write better, more targeted prompts.
-
2. Use other agents as specialists
You act as the ferry between agents.
Ask the Head Agent for a prompt:
“Write a clear prompt for Claude to stress-test this idea.”
Send it to the specialist:
Open a new chat with the specialist agent (Claude, Gemini, Perplexity, Grok, etc.).
Paste the prompt and add any extra context if needed.
Bring the answer back:
Copy the specialist’s response back into the Head Agent.
Tell the Head Agent which agent it came from and what you were asking for.
Ask the Head Agent to:
- summarise it,
- compare it with other inputs,
- and note any follow-up questions or gaps.👉 Repeat this loop a few times if needed (e.g. “get one more angle” or “ask Perplexity to check sources”).
-
3. Head agent synthesises
Once you’ve had enough input, go back to the Head Agent and ask for a clear synthesis:
“Summarise the key points in plain language.
- Suggest 2–3 good options or directions.
- List 3–5 concrete next steps for me.”Optionally, ask:
Draft a short email or doc you can share.
Turn the result into a checklist, outline or plan.
Note any open questions or risks to revisit later.
👉 Your job is to review, edit and decide – not to accept everything at face value.
Next steps
Mental Model 3.0 framework to think about your role as editor / coordinator and how you brief agents.
First 7 messages with a new agent to improve your initial setup with a Head Agent.
Write your own Rails (boundaries & safety rules) so multi-agent work stays safe and human-led.