How to automate tasks with AI (a five-step method)
Not a tool list — a method. The same five steps we use on client work, usable whether you build it yourself or hire it out.
Most "how to automate with AI" guides are tool reviews wearing a how-to costume. Tools change monthly; the method doesn't. Here's the one we run on every engagement — it works identically if you DIY it.
Step 1 — Pick one task (resist the platform urge)
Not "automate my business." One task, chosen by the three-question test: happens often, follows rules you could explain, cheap if it goes wrong once. Write down its trigger and its finished result.
Step 2 — Write the rules as if training a temp
Spend 30 minutes writing how you actually do it: what you look at, how you decide, what the exceptions are, what "good" output looks like (save three real examples). This document is 80% of automation success and requires zero technology. If you can't write it, the task isn't ready — that's a useful discovery, not a failure.
Step 3 — Choose your route
Three options, honestly compared in do I need a developer?: use an AI assistant manually (fine for low volume), build it in no-code tools with AI steps (fine if you'll maintain it), or have it built and run for you (right when it touches customers or money). The rules document from step 2 is the input to all three.
Step 4 — Add guardrails before volume
Whatever the route: an approval step on anything outbound, hard limits on anything financial, logging on everything, and a named human who owns "is it working?" Loosen later, on evidence — never on optimism.
Step 5 — Measure one number
Decide before launch what the automation should move: response time, hours spent, quotes sent per week. Check it at week two and week six. If it moved, scale to the next task; if it didn't, the rules document needs work — the diagnosis is almost never "the AI isn't smart enough."