I was having one of those conversations with AI (yeah yeah, i know) that starts out as a fairly innocent question and somehow ends somewhere much more intriguing. I asked it what it thought were the five jobs AI would never replace. It gave me the usual answers – nurses and doctors, hands on caregivers, skilled tradespeople such as plumbers and electricians, early childhood teachers and childcare workers, therapists and counsellors, and then a broader category of jobs built around trust, relationships and human judgement. The argument was that AI can process information, generate text, analyse data and increasingly perform tasks that once required a person sitting at a desk for hours, but there are things that are much harder to automate particularly physical presence, human relationships, trust and the kind of judgement that happens when the situation in front of you does not neatly resemble anything in the database.
What about leadership? “AI can replace some of the functions of leadership, but leadership itself is much harder to replace,” they said.
Then, somewhere halfway through the conversation, I realised I had been asking the wrong question. Maybe the question isn’t what jobs AI can never do. Maybe it is what happens when AI learns a particular person well enough to do their job. So I paused for a minute, and then plunged head on. Could you run Sunset in Africa?
I was joking…..mostly. The uncomfortable part is that, at this point, this AI does know quite a lot about how I work. In fact, when I asked it what it knew about the way I work, it (they?) came back with an answer that was both flattering and mildly terrifying.
“At this point, I know quite a lot about how you work. I know you don’t like generic AI language. I know when you say ‘let’s write this’ you usually want something flowing, human and journalistic rather than a perfectly symmetrical SEO article. I know you like a provocative opening when the subject calls for it, but you don’t want manufactured outrage. I know you obsess over internal links, search intent, titles, social packaging and whether something actually sounds like Sunset rather than a content farm.”
It was right about some of it’s analysis, especially the titles’ part. I could spend hours obsessing over the perfect title (I’ve come to realise that I have a disease called perfectionism.)
“And?” I pushed.
“So yes, theoretically, if an AI had enough of your historical work, conversations, analytics, editorial decisions, relationships and processes… you could build an AI that runs a frighteningly convincing version of Sunset. 😭”
Was it right? Like i said, It knows I can spend an unreasonable amount of time arguing over a headline, how I don’t like short, choppy writing and that I will immediately notice when the ideas sounds like generic AI copy. It knows I care about search intent, content clusters, meta descriptions and whether an idea is something that people can actually resonate with. Whether this topic or that topic actually matters to my readers rather than simply filling a page. It knows that sometimes I want a straight, journalistic explainer and sometimes I want to wander a little and let an idea breathe. It knows the subjects I keep returning to, from grief and death to health, technology, AI, Education, culture, travel and all those strange little things happening around us that become much more interesting when you look at them from a slightly different angle.
It also knows something that has taken me quite a while to figure out myself: I don’t simply want Sunset in Africa to publish articles. I want it to sound like Sunset…if that even makes sense. A distinction that matters because Sunset has changed considerably from where it began. It started as a blog where i shared my own personal stories and opinions on this and that. Started as a blog where I published fictional stories but over time I have become much more deliberate about what I am building. The stories, the voice, the conversations, the people I meet, the brands I work with and even the things I obsess over in Google Search Console have gradually become part of the publication’s identity. So when I asked the AI whether it could almost run Sunset, it wasn’t entirely ridiculous.
Simply, the answer was yes.
That It could research stories, find potential angles, suggest headlines, build content clusters, analyse what people are searching for, draft and edit articles, write social captions, identify internal links and turn one article into multiple pieces of content for different platforms. It could learn the workflow and the preferences. It could learn the patterns. And that is where the conversation became much more interesting to me, because if a machine can learn how a person works, how much of that person’s work is actually the person?
The AI put it this way: “Imagine I had everything you’ve ever published. I could learn: Winnie tends to notice X → connect it to Y → ask this particular question → structure the article this way → use this tone → link these kinds of stories → package it for Facebook like this. That’s incredibly powerful.” It is hard to argue with that. Give a machine enough examples and enough context, and eventually the thing that once looked like intuition can begin to look like a pattern.
And perhaps that is the part of AI that fascinates me more than the old question of whether it can write an article. Writing is increasingly becoming the easy part. The more interesting question is whether AI can learn the person behind the writing. If you gave it every article I have ever written, every headline I have chosen, every headline I rejected, every conversation about what makes a good story, every analytics report, every editorial decision and enough of the conversations I have had about what I am trying to build, it could start constructing a map of how I think. Winnie sees X, connects it to Y, asks Z, finds the human angle, writes it this way and packages it for this audience. At some point, it could become very good at predicting what I would do next.
But then the AI made an argument that I found much more reassuring. “AI can learn your patterns but it can’t necessarily become the person generating them,” it told me. So, this is where AI gets much more complicated. Because imagine something happens in Kenya that I have never written about before. I see it on my phone one morning and it kinda bothers me. I remember a completely unrelated conversation I had three months earlier, connect the two things and suddenly think, Wait. There is a story here. That connection might not be a formula I have consciously followed before. It could come from experience, memory, curiosity, context, instinct or simply having lived long enough to accumulate a strange collection of unrelated things in my head (sic)
The AI explained it another way: “You can give AI your articles. Your analytics. Your conversations. Your editorial history. But then something happens in Kenya that wasn’t in the dataset. You see it. Something about it bothers you. You remember a conversation from three months ago. You connect two completely unrelated things. And suddenly you think: ‘Wait. There’s a story here.’ That’s where the distinction between learning how someone works and being that person becomes important.” That, I think, is the bit we keep losing when we talk about artificial intelligence replacing people. We tend to reduce human work to the visible output; the article, the photograph, the diagnosis, the spreadsheet and the finished product. But often the valuable part happened before the output existed.
There is also the question of having something at stake. An AI can know that I get excited when an article I least expected suddenly becomes my most read story of the month. It can know that I will obsess over whether a pitch will land, and that I get irritated when analytics refuse to give me a straightforward answer. It can even know that I will probably spend far too long debating whether a word like “affordable” or “cheaper” sounds better in a headline. But knowing those patterns isn’t the same as experiencing them. The AI said something that stuck with me: “It wouldn’t know what it feels like when an article you’ve worked on suddenly becomes your most read story of the month. It wouldn’t get excited because a CEO finally replied to an email. It wouldn’t lie awake wondering whether a pitch is going to land. It wouldn’t meet someone at an event and think, ‘I want to talk to this person again.’ And perhaps most importantly, it wouldn’t have anything at stake.”
I do.
That last part is difficult to dismiss because perhaps the future isn’t actually AI versus humans. Perhaps it is humans who know how to use AI versus humans who don’t. If I can use AI to research faster, identify patterns I might have missed, organise my ideas, analyse traffic, clean up a draft and turn one story into several pieces of content…..well, does that make my work less human? It could actually give me more time to do the things that made me want to build Sunset in the first place: notice things, ask questions, talk to people, make connections and tell stories.
Which brings me back to my original question about the jobs AI will never replace. I am no longer sure there is a useful list of five. There may be jobs that are much harder to automate because they depend heavily on physical presence, trust, relationships, unpredictable environments or emotional connection, but almost every job contains tasks that AI can eventually learn. The more interesting question might therefore be: what remains uniquely human after AI has learned everything else?
And that is a much scarier question for people like me, because the answer cannot simply be, “AI can’t write like me.” It increasingly can. It might even get remarkably good at sounding like me (here is where I pause to drink a tall glass of water).
The answer has to be somewhere deeper than the words on the page. It has to be the curiosity that made me look at something in the first place, the experience that allowed me to recognise a connection, the relationships that gave me access to a story and the personal stake that made me care enough to pursue it.
So could AI eventually help run Sunset in Africa? Maybe. After our conversation, I am much less comfortable saying no. Give it enough information and enough time, and it could probably do an extraordinary amount of the work. It could know how I write, what I obsess over, what kind of stories I chase (trust me, every writer has that which is dear to them), what kind of headlines I like and even what I am likely to argue with it about. It could become incredibly good at predicting what I would do next.
But there is still a difference between predicting what I might do and being the person who wakes up one morning, sees something happening in the world and decides that that matters.
For now, at least, I think I’ll keep that job.
Phew.
