Guide
Turn your SOPs and docs into instructions your AI agent follows
Most small businesses already have the instructions an AI agent needs. They're just buried in SOPs, checklists, old emails and a Notion or Google Drive folder. The work is turning them into short, explicit steps the agent can follow, and giving it read access to the right docs and nothing else. Anthropic's prompting guide puts the test simply: if a colleague with no context would be confused by your instructions, the model will be too (Anthropic docs). Claude Projects and ChatGPT Projects both let you keep instructions and files in one workspace; Notion AI Connectors can pull answers from apps like Google Drive and Slack, with citations, on Business or Enterprise plans (Notion Help). Time figures here are estimates, not measurements.
What to hand to an agent first
Rewrite one SOP as steps
Pick a task you repeat weekly. Rewrite it as numbered steps with the inputs, the output format and when to stop and ask you. Numbered steps help when order or completeness matters (Anthropic docs).
Add the 'why' and the edge cases
For each step, add one line on why it matters and what to do when something is missing. Context and reasons help the model apply the rule correctly.
Put it in a project
Keep the instructions plus the 3 to 5 docs it needs in one project. Claude Projects are self-contained workspaces with their own chat history and knowledge base, and free accounts can create up to five (Claude Help Center).
Connect the knowledge base read-only
In Notion, an API connection can only touch pages you manually share with it (Notion Help), so share one 'Agent docs' page, not the whole workspace. In Google Drive, share a single folder with view access (Google Drive Help).
Test against real cases
Run last month's real examples through it and compare with what you did. Fix the SOP wording, not just the output, when it gets one wrong.
Not sure which jobs fit your business?
Build my free AI plan →Ready-made use cases
- 01Client onboarding on autopilot
When a client pays, the agent sends the welcome pack, intake form and kickoff booking link.
~2h/week · estimate - 0224/7 FAQ support in your voice
The agent answers common customer questions from your docs and hands off anything tricky.
~8h/week · estimate - 03Inbox triage with drafted replies
The agent sorts your inbox into reply / FYI / ignore and drafts replies in your voice for you to approve.
~5h/week · estimate - 04Weekly CEO review & plan
Every Friday the agent compiles wins, numbers, stuck items and proposes next week's priorities.
~2h/week · estimate - 05Personal learning digest
The agent reads the newsletters you never get to and sends the 5 ideas worth your time.
~2h/week · estimate
Example: Grok Bot setup
Grok Bot routine: Client onboarding on autopilot Schedule: On new payment When a new client pays, send my welcome email with intake form and kickoff booking link, create their project page in Notion from my template, and remind me if they haven't booked in 48h. Rules: draft first, never send or post anything external without my approval. Keep summaries short.
Grok Bot, ChatGPT, Claude and n8n versions: all templates
Setup steps
- Pick one weekly task and rewrite its SOP as numbered steps with inputs, output and stop-and-ask rules.
- Add one line of 'why' and the edge cases to each step.
- Create a project with those instructions and the 3 to 5 docs the task needs.
- Share one Notion page or one Drive folder with the agent, read-only, and nothing else.
- Test it on last month's real cases and fix the SOP wording where it goes wrong.
Keep these human
- Share one folder or page, not your whole workspace or drive.
- Keep passwords, bank details and customer personal data out of the docs you connect.
- Mark each SOP with an owner and a last-reviewed date so the agent isn't working from stale rules.
- Anything that sends, pays or deletes stays a draft for you to approve.
- Start in draft mode: the agent prepares, you approve, then you loosen the leash routine by routine.
FAQ
What should an AI agent do first for Turn your SOPs and docs into instructions your AI agent follows?
Rewrite one SOP as steps: Pick a task you repeat weekly. Rewrite it as numbered steps with the inputs, the output format and when to stop and ask you. Numbered steps help when order or completeness matters (Anthropic docs). Add the 'why' and the edge cases: For each step, add one line on why it matters and what to do when something is missing. Context and reasons help the model apply the rule correctly. Put it in a project: Keep the instructions plus the 3 to 5 docs it needs in one project. Claude Projects are self-contained workspaces with their own chat history and knowledge base, and free accounts can create up to five (Claude Help Center).
What should stay human?
Share one folder or page, not your whole workspace or drive. Keep passwords, bank details and customer personal data out of the docs you connect. Mark each SOP with an owner and a last-reviewed date so the agent isn't working from stale rules. Anything that sends, pays or deletes stays a draft for you to approve. Start in draft mode: the agent prepares, you approve, then you loosen the leash routine by routine.
Is there a ready-made template?
Yes. Start with "Client onboarding on autopilot": copy-ready Grok Bot, ChatGPT, Claude and n8n templates are at https://whatshouldmybotdo.com/use-cases/client-onboarding#templates.