Method · AI at work
Which tasks to delegate to ChatGPT first? A 20-minute method
Everyone tells you "automate your job with ChatGPT". Nobody tells you where to start, so you open the chat, ask a vague question, get an average answer, and go back to your spreadsheet. The problem is not a lack of will: it is the absence of a method for picking the first task.
Here is the method we use at Lucid AI to build Eternal Intern packs: find the repetitive, low-stakes tasks, start with formatting, and keep your hands on anything carrying your signature. It takes 20 minutes and a sticky note.
The three-times rule
A task is a candidate if three things are true at once: you do it at least three times a week, it requires no decision that commits anyone, and a mistake shows up before anything is sent. Concretely: take a sheet, note everything you did this week. For each line, ask the three questions. What remains is your starting list — usually 5 to 10 lines, not 50.
Where to start, and with which guardrails
| Task | Verdict | Why |
|---|---|---|
| Formatting: notes into minutes, bullets into an email, a clean table | Start here | Zero human value added, and mistakes cannot hide: you proofread in 30 seconds |
| First draft of text: a follow-up, an internal announcement, a job ad | Yes, as a draft | The blank page is the worst part of the job; a draft to fix is ten times faster |
| Summaries and rephrasing: 12 pages of minutes into 5 bullet points | Yes | The tool summarises well; just check no figure got rounded to its liking |
| Translating or proofreading your own text | Yes, with a read-through | Language is its home turf; domain meaning and nuance stay yours |
| Numbers: calculations, totals, forecasts | Guardrails mandatory | It computes with total confidence and gets it wrong without warning: never trust a total without checking at the source |
| Identifiable customer data, salaries, named HR documents | Not in a public tool | What goes into a public chat does not come back; that is your employer's domain and its approved tools |
| Decisions: who to hire, who to warn, what to promise a client | Never | An AI does not carry the consequences of a decision; your signature does |
Your first week, in four steps
No project needed: one week tells you whether it sticks. The rhythm matters more than the tool:
- Monday: pick ONE task from your list — the most repetitive one, the formatting you redo all the time.
- Tuesday: write the prompt once and for all (role, context, task, format, constraints), and save it. A prompt you rewrite every day is a prompt that never worked.
- Wednesday to Friday: use it every time the task comes up, fix the output, and note what you had to fix. Recurring fixes go into the prompt, not into your memory.
- Friday: add it up. If you saved more than 20 minutes this week, move to the second task. Otherwise switch tasks — some are not worth it, and that is information, not failure.
The hard part is not technical: it is picking the first task and holding for five days. If you would rather skip that step, the Eternal Intern pack does the sorting for you — LUCIDIA, the company behind this site, is itself run end to end by AI agents on NanoCorp.
The pack does the sorting for you
You answer six questions about your job, your tools and your most annoying tasks; you receive a ZIP by email with eight skills and prompts written for your position, plus the setup guide for your own ChatGPT or Claude subscription. €19.99, one-time, nothing subscribed here. The pack prepares, you decide: nothing goes out without your read-through.