Lesson 7 of 8 · 8 min read
“AI automation” sounds like something that requires a developer, a budget, and six months.
For most people it doesn’t. It means noticing that you do the same fiddly thing every Tuesday, and not doing it by hand any more.
This lesson is about finding those tasks in your own job — the ones nobody will ever assign you to fix, because they’re too small to be projects and too frequent to be nothing.
What automation means here
Forget robots. There are three levels, and almost all the value is at the bottom two.
Level 1: a prompt you reuse. You write a good prompt once, save it, paste it in whenever you need it. That’s it. That’s automation — you’ve stopped re-deriving the same instruction every time.
Level 2: a saved setup. Most tools let you save a reusable configuration with your context and instructions baked in, under some product name or other. Now you don’t even paste the prompt. You open the thing and give it today’s input.
Level 3: connected systems. AI wired into your other tools, running without you. Real, powerful, and genuinely a project — usually involving IT, budget, and someone whose job it is.
Beginners assume the value is at Level 3. It isn’t. Almost everyone can get most of the benefit at Levels 1 and 2, this week, with no permission and no spend. Level 3 is where it goes wrong expensively.
Finding the tasks worth it
Not everything repetitive is worth automating. Three questions.
1. Do I do this at least weekly?
Under that, the setup costs more than it saves. Feels productive, isn’t.
2. Is it mostly made of words?
Back to Lesson 1. Drafting, summarising, reformatting, tidying, translating — yes. Anything hinging on precise numbers — read Lesson 6 again first.
3. Does it need my judgement, or just my time?
The distinction that matters. Deciding what to say to an unhappy client needs you. Formatting the update into the same structure you use every week doesn’t.
Tasks that pass all three are your list. For most office workers it’s short and boring: the weekly summary, the same kind of reply to the same kind of email, meeting notes into actions, cleaning up someone else’s formatting.
Boring is the point. Boring and weekly is exactly what you want to stop doing.
Try this: the two-week audit
Open a note. For two weeks, every time you catch yourself thinking “I’ve done this before” — write it down. One line.
That’s it. No analysis, no system.
At the end you’ll have a list, and it will surprise you. Not because it contains things you didn’t know you did — because seeing them written together makes obvious how much of your week they add up to. Most people can’t name their repetitive tasks from memory. They’re invisible precisely because they’re routine.
The top three on that list are where you start.
Building a reusable setup
Say you write a weekly team update. Same shape every time, different content.
The manual way: stare at the blank page, remember roughly what goes in it, write it.
Level 1: save a prompt with the structure, the tone, the audience, and an example of a good past one. Paste it in with this week’s notes. Two minutes.
Level 2: put that prompt into a saved setup — every tool has a version of this — along with a few past updates as examples. Now you open it and paste this week’s notes. Thirty seconds.
Notice what’s actually happening: you’re doing Lesson 4 once instead of every week. Context, task, format, example — written properly one time, then reused forever. Automation here is just prompting that you stopped repeating.
That’s why the earlier lessons come first. Without them you’d be saving a bad prompt and automating a mediocre output.
What not to automate
Anything where being wrong is expensive. Client-facing commitments, legal or financial claims, anything with your name on it that can’t be taken back. Draft with AI, absolutely. Don’t ship without reading.
Anything requiring context you can’t paste. If doing it well depends on knowing that this client was difficult last year and the account manager is on thin ice — that lives in your head, not in the prompt.
Anything that’s the actual work. If you’re a copywriter, don’t automate the copy. Automate the brief-summarising, the version-tracking, the reformatting. Automating your craft is how you become worse at it while looking faster.
Anything you do rarely. The setup won’t pay back. Just do it.
The trap worth avoiding
The biggest mistake isn’t automating badly. It’s automating something that shouldn’t exist.
That weekly report nobody reads. The status update duplicating what’s already in the tracker. The formatting that only exists because someone in 2019 liked it that way.
AI makes it very easy to keep producing these forever, faster, with less friction — which removes the one thing that might eventually have killed them. Nobody questions a task that only takes thirty seconds.
So before you automate, ask what happens if you just stop. Sometimes the answer is nothing. That’s a better outcome than any automation, and you’ll never find it if your first instinct is to make the pointless thing efficient.
Start with one
Not a system. One task.
Take the top item from your audit. Write a good prompt for it using Lesson 4. Use it for a fortnight. Refine it when it’s not quite right.
Then, and only then, do the next one.
The failure mode here is enthusiasm — trying to automate your whole week in an afternoon, building ten mediocre setups you don’t trust, and quietly going back to doing it by hand. One thing that works is worth more than ten that nearly do.
Key takeaways
- Most of the value is in reusable prompts and saved setups — not connected systems
- Worth automating: weekly or more, made of words, needs your time but not your judgement
- Run a two-week audit — write down every “I’ve done this before”
- A saved setup is just Lesson 4 done once instead of every week
- Don’t automate your actual craft, or anything expensive to get wrong
- Ask whether the task should exist at all before making it efficient
- Start with one thing and get it right
Next: Lesson 8 – Ethics, Limits, and Failures
Last lesson. Everything you’ve learnt assumes you know where the edges are — so that’s what we finish with.