Two years of confident predictions about AI and IT jobs have produced a reasonably clear picture, and it does not match most of the predictions.
Here is what actually shifted, drawn from what teams are doing rather than what analysts forecast.
What Shrank
Not roles, tasks. That distinction explains most of the confusion.
Tier-one ticket handling shrank, with organisations reporting AI resolving a substantial share of routine issues without human involvement. First-draft code writing shrank as a share of a developer day. Routine data extraction and report assembly shrank. Basic content production shrank dramatically.
What did not happen is entire roles vanishing. What happened is that the easy 60% of several jobs got automated, leaving the hard 40% and a lot of new supervision work.
The Roles Genuinely Growing
| Role | What it does | Where it comes from |
|---|---|---|
| AI platform engineer | Runs the gateway, identity, budgets and observability | Platform and DevOps |
| Evaluation engineer | Builds and maintains golden sets and quality pipelines | QA and data |
| AI governance lead | Owns policy, inventory and compliance | Security and risk |
| Agent operations | Monitors agent fleets and handles exceptions | Service desk leads |
| Data quality engineer | Keeps the pipelines that AI depends on honest | Data engineering |
| Integration specialist | Connects systems so agents can reach them | Middleware and IT |
Notice what these have in common. Every one requires knowing how your organisation actually works. That is not a skill you can hire cheaply from outside, which is why internal moves fill these roles more often than external hires.
The Skill That Rose Most
Judgement about when something is quietly going wrong.
When agents handle the routine cases, the remaining work is exceptions and anomalies. The valuable person is the one who notices that resolution times look fine but reopen rates crept up, or that the model has been confidently wrong about one customer segment for a fortnight.
That skill is unevenly distributed and hard to teach quickly. It is also what makes experienced staff more valuable rather than less, which is the opposite of the common prediction.
What This Means for Hiring
- Hire for judgement over tool familiarity. Tools change every six months. Judgement does not.
- Promote internally into AI roles. Domain knowledge is the scarce half, and the AI half is learnable.
- Stop hiring for pure production work. Roles defined by volume of output are the exposed ones.
- Invest in your junior pipeline anyway. If the industry stops training juniors because AI does the easy work, the senior shortage in five years will be severe and self-inflicted.
The Uncomfortable Part
The traditional route into IT ran through the tasks that got automated first: tier-one support, basic scripting, simple content and reporting. That ladder lost several rungs, and nobody has agreed on what replaces them.
Organisations that solve this deliberately, by pairing juniors with agents and teaching supervision early, will have a hiring advantage in a few years. Most will not solve it, and will complain about the talent shortage they helped create.
Conclusion
Tasks shrank, roles changed shape, and the growing jobs are supervision, evaluation, governance and integration. Hire for judgement rather than tool lists, promote from inside because domain knowledge is the scarce ingredient, and keep investing in juniors even though the obvious training tasks disappeared. The teams that keep their experienced people and give them better instruments will do considerably better than the ones that treated AI as a headcount exercise.
Frequently Asked Questions
Should junior developers be worried?
They should be strategic. The work of typing first drafts is commoditised. Reviewing, debugging, integrating and understanding systems end to end is not, and that is where to build depth quickly.
Which certifications are worth it?
Fewer than vendors suggest. Demonstrable work, an evaluation set you built, an agent you shipped and supervised, tends to carry more weight than a certificate in a tool that will change.
Is the IT job market shrinking overall?
It is redistributing more than shrinking. Production roles are under pressure while supervision, governance and integration roles are growing, and the second group is currently harder to fill than the first is to cut.