🙅 7 Jobs I Never Do Myself Anymore

There’s a question I ask about every job that crosses my desk now, and it’s changed how I work more than any tool, app or method from my six years building the back ends of other people’s businesses: is this a first-pass job or a final-say job? First-pass jobs are about getting from nothing to something - a blank page to a draft, a forty-page PDF to a summary, a jumble to a list. Final-say jobs are about judgement, taste and consequences. The realisation that quietly reorganised my working life: AI is brilliant at first passes and I was doing all of mine by hand, like someone hand-washing clothes while stood next to their own washing machine.

So over the past couple of years, seven jobs have moved permanently onto what I call my Never List - jobs I never start from scratch myself anymore. Not “sometimes, when I remember”. Never. Here they are, each with the actual prompt I use, plus - because these lists are dishonest without it - the jobs I’ve deliberately kept.

I do all of this in Claude because that’s where I live, but every one of these works in ChatGPT too.

1. First drafts of anything

The job. Emails, guides, proposals, captions, difficult paragraphs of any species. I haven’t faced a blank page in a very long time and my quality of life has improved accordingly.

My prompt:

Nick this one, fill in the blanks:

First draft. The thing: [what it is - email, post, one-pager]. It's for: [who'll read it]. It needs to achieve: [the actual goal - "get them to confirm the date", "explain the delay without grovelling"]. Rough thoughts to build from: [everything in your head about it, unsorted].

Write it how I'd say it out loud on a good day - warm, plain, no corporate filler. Mark anything you had to guess at with [CHECK].

Worth knowing. The “rough thoughts” line is what makes the draft yours instead of generic - even three messy fragments change everything. And [CHECK] marks are the safety net: the draft flags its own guesses instead of smuggling them past you.

2. Summarising long documents

The job. Contracts, reports, terms and conditions, policy updates, anything described as “please see attached” with a straight face. I read summaries first and originals second, and only the parts the summary tells me to.

My prompt:

Nick this one:

Summarise this for someone busy: [paste or attach it]. Give me the gist in 5 bullet points, anything that requires action or a decision from me (with deadlines), anything unusual, risky or different from standard, and the 2-3 sections worth reading in full myself. Tell me what kind of document this actually is if it isn't obvious.

Worth knowing. That last request - which bits to read in full - is the difference between a summary that replaces reading (risky for anything that matters) and a summary that targets your reading (genuinely safe). Ten minutes on the two clauses that matter beats forty on the whole thing.

3. Turning rambles into tidy notes

The job. After a call, a walk, or one of those shower thoughts that arrives fully formed and evaporates by lunch: I talk (voice mode on the phone) or type a stream of consciousness, and it comes back as notes a functioning adult might have taken.

My prompt:

Nick this one:

I'm going to ramble. Turn it into tidy notes: decisions made, tasks (mine vs other people's), open questions, and ideas worth keeping. Anything I said twice, treat as important. Don't add things I didn't say. Here goes: [ramble away]

Worth knowing. “Anything I said twice, treat as important” came from noticing my own patterns - repetition in a ramble is your brain underlining. And “don’t add things I didn’t say” keeps the notes honest; without it, tidying can drift into inventing.

4. Comparing options

The job. Two software tools for a client. Three flight routes. Insurance quotes, phone contracts, which supplier. Anywhere the information sits in front of me and the decision still won’t form.

My prompt:

Nick this one:

Help me decide. The options: [paste the details of each - screenshots, quotes, links, notes]. What actually matters to me, in order: [e.g. total cost, ease of leaving, how much faff day-to-day]. My gut currently says: [name it, or "no idea"].

Compare them against MY priorities in a short table, flag anything hiding in the small print, then recommend - and tell me what would need to be true for the runner-up to win instead.

Worth knowing. Telling it what my gut says is deliberate. If the recommendation matches my gut, I move fast with extra confidence. If it doesn’t, that mismatch is exactly the conversation worth having. Either way the decision that used to marinate in open tabs for a fortnight takes ten minutes.

5. Prep research

The job. Before a call with someone new, a session I’m teaching, a decision about a tool - the background-gathering that used to eat a morning. Say I’ve got a call with a potential client who runs an online course business: I want the landscape, the obvious questions, the thing I’d look silly not knowing.

My prompt:

Nick this one, fill in the blanks:

Prep me for: [the call / meeting / decision]. Context: [who's involved, what it's about, what I want out of it, anything I already know].

Give me: the background I should have (use web search where it helps), the 5 questions I should ask, the 2 questions THEY will probably ask me and strong answers, and one thing that would make me sound like I'd done my homework - because I will have. This briefing counts.

Worth knowing. Prep research is where AI making things up will bite hardest, so anything factual that will come out of my mouth in the meeting gets a source or a quick check first. The shape of the briefing is reliably excellent; the individual facts are trust-but-verify.

6. First passes at tricky messages

The job. Chasing an unpaid invoice without souring the relationship. Saying no to work. The “I think there’s been a misunderstanding” email. Anything where I’d otherwise draft eleven versions in my head at 2am.

My prompt:

Nick this one, fill in the blanks:

I need to send a tricky message and I keep putting it off. Situation: [what happened, honestly - including your part in it if any]. The person: [who they are and what the relationship is worth to you]. What I need to happen: [the outcome]. What I'm afraid of: [sounding harsh, being a pushover, losing them - name it].

Draft it: honest about the issue, warm to the person, clear about what happens next. Short. Then give me one line on why this framing works, so I stop second-guessing it.

Worth knowing. “What I’m afraid of” is the line that changes everything, because tricky messages go wrong when the fear writes them - fear of being harsh produces mush, fear of being a pushover produces harsh. Naming the fear lets the draft steer around it. This job alone justifies the whole Never List; 2am me agrees.

7. Naming things

The job. Workshops, newsletters, offers, projects, guides like this one. Naming feels tiny until you’ve lost forty minutes staring at a document called “Untitled 7”.

My prompt:

Nick this one:

Name this: [describe the thing, who it's for, and the feeling it should give]. Give me 15 options across the range: safe and clear, warm and human, playful, and 2 that are probably too much (they help me see the edges). No puns unless they're genuinely good. Then tell me your top 3 and why.

Worth knowing. Ask for fifteen, never five. The first five are always the obvious ones you’d have thought of in the shower; the good stuff lives in options eight to fifteen, once the obvious is used up. And the “too much” pair genuinely helps - seeing the edge of the map tells you where the middle is.

The jobs I kept

Now the honest bit, because a list like this can read like I’ve automated myself out of my own life. The Never List is first passes only. The final-say jobs stayed mine, on purpose: anything with my name on it gets read, reshaped and approved by me before it goes anywhere - every guide, every client email, every message a human will receive thinking it came from a human. The judgement calls stayed too: what to charge, who to work with, what to say no to. AI informs those decisions; it doesn’t get a vote. The line I hold is simple: AI does the jobs where speed matters, I do the jobs where meaning it matters. The day I can’t tell which is which, I’ve been doing this wrong.

The catch (there’s always one)

Handing over first passes doesn’t hand over responsibility - if a summary misses a clause or a draft gets a fact wrong and I send it anyway, that’s on me, not the machine. Everything here still needs your eyes before it acts on the world. And fair warning: the first week of a Never List feels slower, not faster, because you’re learning to brief instead of do. It flips somewhere in week two. Push through the awkward bit.

If today only has five minutes in it

Move job number one - first drafts - onto your own Never List today. It’s the job you do most often, the prompt takes a minute, and once starting-from-nothing leaves your life, you’ll wonder what else can go.

For the household versions of these - the confusing letters, the fridge, the gift panic - 5 Boring Jobs You Should Have Handed To AI Already is the companion piece to this one.