How to Write the Rulebook for Your AI Agent Team
Once a small or medium business has a solid foundation, meaning a clear website, reliable lead capture and basic search set-up, the next step is often an agent team: AI helpers that handle routine work every day. The difference between a team that helps and one that causes problems is rarely the technology. It's the rules they work to.
Here's how to write a rulebook your agent team, and the people around it, can follow.
Why rules come first
A new employee gets an induction, a job description and someone to ask. AI helpers need the same, written down. Without written rules, each helper makes its own guesses about tone, facts and limits. With them, work is consistent, mistakes are easier to trace and you can trust the team with more over time.
1. Define each role
Give every helper one clear job. For example:
- Content: drafts articles and posts from an approved topic list
- Leads: sorts new enquiries and prepares replies
- Reporting: summarises what happened each day or week
- Coordination: keeps track of tasks and flags anything stuck
For each role, write what it does, what it hands to others and who it reports to. Overlapping roles lead to duplicated or conflicting work.
2. Name the source of truth
List where the correct facts live: your website, your price list, your policies, your product details. Helpers must check facts there, not invent or guess them. If a fact isn't in the source of truth, the rule is to ask, not to fill the gap.
3. Write a "never do" list
This is the most important page. Typical entries:
- Never invent reviews, clients, results or figures
- Never promise prices, discounts or delivery dates that aren't confirmed
- Never make claims the business can't prove
- Never share customer details outside their own order or enquiry
- Never spend money or change billing
- Never publish or send in the owner's name without approval
Keep it short and specific, so it can be checked easily.
4. Set approval gates
Decide which actions need a person's yes before they happen. Common gates include publishing content, sending messages to customers, launching or changing ads, and anything involving payments. Everything else, such as drafting, research and internal reports, can usually run without waiting.
Write down who approves each type of action and how quickly they'll respond, so work doesn't stall.
5. Describe the daily routines
List what should happen every day or week, and when. For example: check new enquiries each morning, draft today's content, prepare a weekly summary on Friday. Routines turn the team from something you prompt into something that runs.
6. Build in quality checks
Every output should be checked before it goes out, either by another helper, by a person, or both. Useful checks include:
- Do facts match the source of truth?
- Does it break anything on the "never do" list?
- Does it repeat work already done?
- Do the links work?
When a mistake slips through, add a rule so it doesn't happen again. The rulebook gets better with use.
7. Keep it short and living
A rulebook nobody reads is useless. Aim for a few pages, written in plain language, with the most important rules first. Review it monthly. Remove rules that no longer apply, and add new ones as the team takes on more work.
A starter template
- Business summary: what you sell, to whom, where
- Roles: one paragraph per helper
- Source of truth: where the facts live
- Never do: the short list
- Approvals: what needs a yes, and from whom
- Routines: daily and weekly jobs
- Quality checks: what's checked before anything goes out
Ready to upgrade to an agent team?
Apne Technologies builds AI-run business infrastructure for small and medium businesses worldwide, foundation first, delivered remotely. An agent team with written rules, daily routines, human approval and quality checks is the next step: we run our own work this way today, and it's available as an upgrade for existing clients. Learn more at apnewebsite.com, or message us on WhatsApp to talk about your business.