AI-run operations

First principle

Automate only what has been validated. Operations means repeatable delivery plus support plus measurement, increasingly run by AI agents. Do it manually first to learn the SOP, then hand the SOP to AI. Premature automation just scales the mistakes too.

Key steps

1. 1

   Do it manually, write the SOP

   Deliver by hand a few times first and write down every step into a repeatable SOP. Do not freeze a process you have not yet proven.

2. 2

   Find the repetitive, rule-based parts

   In the SOP, circle the parts that are repetitive and rule-based — those are the ones suited to AI.

3. 3

   Hand those to AI

   Hand those parts to AI: support drafts, customer onboarding, reporting. AI brings execution; you bring judgment.

4. 4

   Keep a human check on judgment and edge cases

   Keep a human in the loop wherever judgment or edge cases are involved. AI runs the routine; humans handle the exceptions.

5. 5

   Track delivery quality and time saved

   Keep measuring two things: whether delivery quality holds, and how much of your time is saved per customer.

Copy-ready prompts

I currently deliver to customers by hand. The process is roughly:
[describe step by step what you do from taking the order to finishing delivery]

Help me:
1. Turn it into a clear, repeatable SOP, written step by step.
2. Mark each step as either "repetitive and rule-based, fit for AI" or "needs judgment, must stay human."
3. Point out which step is most worth automating first, and why.

I want to hand this step to AI: [step, e.g. support replies / onboarding / weekly report].

Context:
- My product: [one-sentence product]
- Common cases: [list a few typical scenarios]
- Exceptions that must stay human: [list exceptions]

Design this AI step for me: what context to give the AI, what it outputs, at which point it hands exceptions back to me, and how to check the quality of its output.