I have a simple rule before I use AI to automate anything: Do it manually 3 times first.

—

Most AI automation projects don’t fail because of AI.

They fail because there was no clear process to automate.

I think there’s a fundamental misunderstanding about what AI can do right now. We don’t have an autonomous AI that produces excellent work out-of-the-box now. Because LLMs are an averaging technology. They produce average results by training.

To get great output, we need to orchestrate it with workflows and prompts, crafted by a domain expert. To put it bluntly, what we have is workflow automation 2.0 - better than old Zapier chains, but the same core idea. AI is like a junior employee that needs a clear process to follow.

When people throw a vague task at AI prompt it to ‘act as [expert]’ and it doesn’t work, they blame the tool. “AI doesn’t work.”

The reality usually is they didn’t have a process. They didn’t know how to do it in the first place. AI just made that visible.

I have a simple rule before I automate anything:

Do it manually 3 times first.

• First time - it’s messy. You’re figuring out what needs to happen. Half the steps are wrong.

• Second time - you find a reusable flow. You skip the dead ends from round one.

• Third time - you find the process. The sequence is clear. Inputs and outputs are defined. You know what “good” looks like.

Then write it down and give it to AI.

I use this for everything - from automating content workflows to building AI coding agents. If I can’t do it manually 3 times and write a clear process, I don’t automate it.

This is probably why most corporate AI automation projects stall. They skip straight to “let’s use AI for this” without anyone on the team being able to do the work end-to-end themselves.

Next time someone proposes an AI automation project, ask one question first: “Can you show me the manual process, written down, step by step?”

AI can execute a process. It can’t discover one for you.

#AI #AIAutomation #Founders #LeanStartup