Most people haven’t started with AI for a sensible reason: every explanation begins with vocabulary. Tokens, prompts, context windows, agents. You don’t need any of it. You need the same thing you needed to learn a spreadsheet: one real problem, low stakes, repeated often.
Pick a boring problem you already have
Not “transform my business”. Something small that mildly annoys you every week: the email you dread writing, the meeting notes nobody tidies, the rota, the spreadsheet formula you always google. Boring is perfect — you know what a good result looks like, so you can judge the machine’s work instantly.
Talk to it like a colleague, not a search box
The single biggest unlock: stop typing keywords, start giving context. Paste the whole email thread and say what outcome you want, what tone, what constraints. The model builds its answer from your words — thin words in, thin answer out. Rich context in, and it starts to feel spookily useful.
Don’t learn AI. Do your own job with AI standing next to you.
The ten-minute loop
- Ask — hand it today’s version of your boring problem, with full context.
- Check — read the answer like a manager reviewing a keen new hire. It’s fast and confident; it is not always right.
- Push back — “shorter”, “less formal”, “you missed the part about the invoice”. The second draft is where it earns its keep.
- Repeat tomorrow — the skill isn’t prompting, it’s calibration: learning what it’s brilliant at and where it quietly falls over.
What not to worry about yet
Which model is “best” (pick any big one; move later). Whether you’re prompting “properly” (there is no properly). Building agents, automations, integrations — all real, all later. And don’t paste in passwords, client secrets or anything you’d regret leaving in a taxi; sense over fear, but sense first.
Two weeks of the ten-minute loop beats any course we’ve seen. And once the machine stops feeling like magic and starts feeling like a slightly overconfident colleague, you’re no longer “getting started with AI”. You’re just working.