What jobs will AI create rather than destroy?

The honest framing is that tasks get automated, not whole jobs — and roles change shape rather than vanishing. The jobs most affected are ones where a large share of daily work is routine information processing; the roles that grow are ones that direct, verify, integrate or take responsibility for that work. Work that's growing directly: - **People who build and deploy AI systems** — ML engineers, data engineers, and increasingly 'AI application' developers who assemble models, retrieval and tools into products. Enormous demand, and it requires ordinary software skills more than a research background. - **Evaluation and quality roles** — deciding whether model output is actually correct and safe in a specific domain. This barely existed five years ago. - **AI governance, risk and compliance** — regulation is arriving, and organisations need people who understand both the technology and the obligations. - **Domain experts who can direct AI in their field** — a lawyer, radiologist, accountant or teacher who uses these tools well is far more valuable than either the tool alone or the professional who refuses to touch it. This is the largest category and the most accessible one. - **Data work** — the models are only as good as the data pipelines behind them, and that plumbing is stubbornly human. Work that stays relatively safe: - **Physical and skilled trades.** Electricians, plumbers, nurses, technicians. Robotics is nowhere near the dexterity and adaptability of a human in an unpredictable physical space, and it's capital-intensive where labour is not. - **Work requiring accountability.** Someone has to be responsible when it's wrong — legally, medically, financially. Responsibility can't be delegated to a model. - **Genuine human relationship work** — care, therapy, teaching, negotiation, sales, management. People want a person, and often the person *is* the product. - **Novel problem-solving in messy real-world contexts**, where the problem isn't well-specified and the data doesn't exist. What's genuinely most exposed: routine writing, basic translation, first-line support, simple data entry and processing, entry-level analysis, and stock content creation. Not gone — compressed, with fewer people doing more. The practical strategy, whatever field you're in: be the person who uses these tools well within a domain you actually understand. The competition is rarely 'AI versus you' — it's 'a person using AI versus a person not using it'. And invest in the parts of your work that involve judgement, responsibility, relationships and physical presence, because those are where the value concentrates as the routine parts get cheap.