Automating the busywork you hate
The biggest wins from AI at work aren't from flashy projects — they're from quietly eliminating the ten tasks you do every week on autopilot. Here's how to find them and build prompts that handle them reliably.
What "automation" actually means here
Automation in this context doesn't mean writing code or connecting APIs. It means building a reusable prompt — one you save, return to, and use repeatedly — so you never have to think about how to phrase a particular request again. The first time you write a meeting summary takes five minutes. The fiftieth time, with a saved prompt, takes thirty seconds.
The goal is a "delegate and go" mindset: treat AI like a reliable assistant you can hand tasks off to, rather than a tool you have to configure from scratch each time.
Finding tasks worth automating
Not everything is worth automating. The best candidates share three qualities:
- Repetitive. You do it regularly — weekly, daily, or with every project of a certain type.
- Formulaic. The structure is consistent even if the content varies. Meeting notes, status updates, job descriptions, reply emails, social captions — all follow patterns.
- Low stakes enough to review quickly. If checking the output takes longer than doing the task yourself, the automation isn't worthwhile yet.
Common automation wins: meeting notes and action items, summarising long documents, writing first drafts of repetitive messages, converting raw notes into formatted reports, categorising feedback or responses, generating options to choose from rather than starting blank.
Building a reusable prompt
A reusable prompt is just a prompt with variables — placeholders you swap in each time you use it. Here's the structure:
- Fixed instructions — what the task is, the format you want, the tone, any constraints. These don't change.
- Variable input — the specific content this time. Often indicated with something like [PASTE MEETING NOTES HERE] or [INSERT CLIENT NAME].
Example: "You are summarising meeting notes for a professional team. Extract: (1) key decisions made, (2) action items with owners and deadlines, (3) any open questions. Format as a bulleted list under each heading. Keep each bullet under 20 words. Meeting notes: [PASTE HERE]"
Save this somewhere you'll actually return to — a notes app, a browser bookmark, a document called "My AI templates". The specific location doesn't matter; the habit of saving does.
The quality-control step
Good automation includes a review step. Don't send AI-generated output without reading it. Your review should be quick — scanning for errors, tone issues, or anything that sounds off — not a full rewrite. If you're rewriting more than 20% of the output, the prompt needs improving, not the draft.
Start small. Pick the one repetitive task that would save the most time this week. Build the prompt for that. Don't try to automate everything at once — you'll build nothing usable. One good template compounding over months is worth more than ten half-finished ones.
Build one reusable prompt for a recurring task
Choose something you actually do regularly. Write a prompt template for it that you can reuse without editing the instructions each time.
Examples: writing meeting summaries, drafting update emails to stakeholders, summarising articles or reports, writing responses to common customer questions, generating social media captions for blog posts.
What format should the output be in? What tone? What constraints apply every single time? Write these as permanent parts of the template.
Decide where the changing input goes. Mark it clearly — [PASTE CONTENT HERE] or [INSERT RAW NOTES]. Keep the variable section clearly separated from the instructions.
Run the prompt with actual content from this week. Note anything that came out wrong, then adjust the fixed instructions to prevent it next time.
What to remember
- Automation = a saved, reusable prompt — no code required.
- Best candidates are repetitive, formulaic, and quick to review.
- Structure: fixed instructions + variable placeholder. Save it somewhere you'll actually use.
- Always review output before sending. If you're rewriting more than 20%, improve the prompt.