I build the systems that take work away from people. So when someone asks me this, I try to give them something more useful than reassurance — and more useful than panic.
The honest answer
AI is not replacing your job. It is replacing specific tasks inside your job — and the ratio matters enormously.
If 80% of your working week is routine, rules-based tasks, your role is going to change significantly, and quite soon. If 20% is, you are about to lose the most tedious fifth of your week and probably enjoy your job more.
So the useful question is not "will AI replace my job." It is: what proportion of what I do is genuinely routine? Answer that honestly and you have your actual risk assessment.
What actually gets replaced
From what I actually build, the tasks that automate cleanly share three properties: clear rules, structured or semi-structured input, and a verifiable output.
- Moving data between systems. The single most common thing I automate. Nobody should be a human copy-paste function.
- First-line triage. Sorting, categorising and routing incoming requests.
- Standard document handling. Reading invoices, extracting fields, filing them.
- Routine reporting. The same report, same shape, every week.
- Repeat questions. The same five answers, delivered sixty times a month.
- First drafts. Of emails, summaries, descriptions — with a human editing.
Notice what these have in common. They are the parts of a job that people describe as "admin" — the parts almost nobody enjoys and nobody was hired for.
What does not
Equally from experience, here is what I repeatedly fail to automate well, and advise clients not to try.
- Anything needing context that was never written down. Knowing that this particular client is difficult in January, or that the invoice from that supplier is always wrong in the same way. This knowledge lives in people, not systems.
- Judgement with real consequences. Not because a model cannot produce an answer, but because someone has to be accountable for it.
- Genuine relationship work. Difficult conversations, negotiation, trust built over years.
- Deciding what should be done at all. AI is good at executing a defined task. Working out which task is worth doing remains stubbornly human.
- Physical work in unpredictable environments. Still much harder than knowledge work, despite the attention going elsewhere.
- Anything where being wrong is unacceptable and unverifiable. If you cannot check the output, you cannot safely automate it.
A pattern worth noticing: AI handles the part of the job you could write instructions for. It struggles with the part you learned by doing the job for three years.
The real risk is different
Here is the thing I think most coverage of this gets wrong.
The realistic near-term risk is not that a system replaces you. It is that a colleague who uses these tools well does the work of two people, and the organisation needs fewer of you.
That is a meaningfully different problem, and it has a meaningfully different response. You do not have to out-compete the technology. You have to not be the last person in your team still doing everything by hand.
That is a far more achievable goal, and it is entirely within your control.
What to actually do
Concrete, in order of effort.
1. Audit your own week
For one week, log what you do. Mark each item routine or judgement. The ratio is your honest position — better than any think-piece, including this one.
2. Automate your own routine tasks first
Before anyone does it for you. Two benefits: you get the hours back, and you become the person who understands how this works. That reputation is worth more than the time saved.
3. Move deliberately toward the judgement work
Whatever part of your role requires context, relationships or accountability — do more of it. That is where durable value is concentrating.
4. Learn the tools well enough to be useful
Not to become an engineer. Enough to look at a process and say "that could be automated, and here is roughly how." In most organisations that person does not currently exist, and the gap is wide.
5. Become the translator
The scarcest role right now is not someone who can build automations. It is someone who understands the business process and the tooling well enough to connect them. Technical people do not know the process. Process people do not know the tools. Being both is rare and paid accordingly.
Why this is not a panic
I say this as someone with an obvious commercial interest in automation, so weigh it accordingly — but I think the honest read is calmer than the coverage suggests.
Every business I have automated for has ended up with the same people, doing different work. The invoicing person stopped processing invoices and started managing supplier relationships. The support person stopped answering "where is my order" and started handling the complicated cases properly, with time to do it well.
That is not a universal law and I would not pretend it is. Some roles genuinely shrink. But the dominant pattern I see up close is redistribution rather than elimination — and the people who came out ahead were the ones who engaged with it early rather than waiting to see.
The window where being early is an advantage is open now. It will not stay open indefinitely, but it is open, and that is a better position than most coverage of this subject would have you believe.