The headlines saying AI is about to delete IT are wrong in the general case and mostly right in one narrow one. Most of the discipline is more or less fine. Some parts are quietly stronger than they were three years ago. And one specific role is in genuine, ongoing structural decline. If you're picking a path right now, the decision worth making isn't whether AI is coming for IT. It's whether you're aiming at the one role it actually is coming for.
Who this is for
- Career changers about to pick a direction and reading enough panic content to wonder if any of this is a sensible bet.
- People already in IT, trying to decide whether to specialise deeper in their current lane or pivot to something more AI-resilient.
- Managers and team leads quietly trying to work out which headcount to defend and which to let attrition handle. (You know who you are.)
The role that is, actually, in decline
First-line, scripted, scope-limited support work. The L1 contact centre tier whose job is to read a runbook, ask three qualification questions, and either resolve a known issue or escalate. That role has been quietly compressing for fifteen years thanks to better self-service and better runbooks. LLMs have accelerated it sharply since around 2023 because they finally do the runbook-following part well enough to deploy.
This isn't a "might happen by 2030" claim. It's happening now in large outsourced contact centres and in the support tiers of SaaS companies. The job isn't disappearing in a single quarter, but the trajectory is unambiguous and it isn't reversing. New entrants into pure L1 scripted support should plan for the role to be smaller every year, not larger.
Note the specificity. Not all helpdesk. Not all support. Not sysadmin, not desktop, not field service. Scripted, scope-limited L1 only. The rest of frontline IT, the bit that requires judgement, physical access, multi-system context, or pattern recognition across an estate, is fine for the foreseeable future.
The roles that are quietly stronger
Cloud and platform engineering. Demand has gone up, not down, partly because every team now wants to bolt some sort of AI-flavoured workload onto the estate and partly because the last few years of cloud sprawl finally need somebody to clean up. The people doing the cleanup are reliably overemployed.
Security generally, and detection engineering and AppSec specifically. Attack surface keeps expanding, regulatory pressure keeps tightening, and AI tooling has made low-effort attacks easier to scale, which means more alerts, more triage, more tuning, more code review. The career remains a slow on-ramp and the entry-level squeeze is real, but the work itself is getting more, not less.
Networking, in the boring on-prem and hybrid sense. Out of fashion in the trade press for a decade. Quietly indispensable. The number of competent network engineers under forty in the UK is small enough that the ones who exist are choosy about who they work for. AI hasn't changed that and isn't about to.
Anything that touches identity. IAM, SSO, privileged access, directory services. Boring on the surface, central to every modern architectural conversation. Salaries have crept up steadily and show no sign of changing direction.
What people get wrong
The first one is confusing "AI can do part of my job" with "AI will take my job." It usually can do part. It usually doesn't take the whole. The actual effect, in most IT roles, is that the boring part of the work compresses and the judgement part expands. The total amount of work mostly doesn't fall. It shifts up the value chain. Same pattern as every previous wave of automation, more or less.
Then there's the assumption that "AI-related work" is the safe bet. It isn't, especially not at the prompt-engineering and chatbot-glue end of it. Most of that work is fashion-cycle adjacent and the skills go stale fast. The durable AI-related careers are mostly in the infrastructure underneath. GPU networking, MLOps, model-serving platforms, data engineering. Those are genuinely hard to break into without a strong systems background. Career changers who skip to "AI engineer" without the systems base mostly don't land.
And the lazy one: treating panic content as data. The volume of thinkpieces about AI displacing IT isn't correlated with the actual labour-market numbers, and is heavily correlated with the ad revenue of the publications producing them. The actual signals (junior hiring rates, senior salary trends, vacancy durations) are mixed but mostly normal. Read the data, not the commentary.
If you're currently in the role that is shrinking
Don't panic, and don't quit IT. The skills are still useful. The specific seat is the problem. Two reasonable moves.
The first is sideways into a role where physical, contextual or judgement work is the centre of the job. Desktop and field support at any decent-sized employer. T2 support at a SaaS company. Junior sysadmin or junior cloud support somewhere cloud-heavy. None of these are immune to automation forever, but all of them have a much longer horizon than scripted L1, and the salary trajectory is meaningfully better.
The second is upward into the parts of L1 that aren't going away. The bits that involve writing the runbooks, training the new starters, or owning the relationship with a specific problematic application that nobody else wants to learn. The people in your team who do those things are not on the list when the automation arrives. The people who only read scripts are.
When to walk away from the plan
If you're currently picking a first IT path and pure scripted L1 is the only realistic option in your local market, take it anyway, but treat it as a strict twelve-to-eighteen-month stepping stone rather than a career. The skills it builds are still real. The seat is just temporary. Plan the sideways move from day one. Read the helpdesk piece for the longer version of that plan.
If you're already mid-career in cloud, security, networking or infrastructure and you're worried, the honest answer is that you're probably worrying about the wrong thing. Pick one part of the AI stack adjacent to your current work (model serving for cloud people, prompt-injection and model security for security people, GPU networking for network people) and learn a sensible amount about it. Don't pivot. Augment.
Where this connects on POST
The pathways page tags each route with an AI-resilience score derived from the work itself, not the marketing. Useful when picking between two otherwise similar lanes. The helpdesk piece is the most relevant adjacent essay for anyone whose specific worry is being on the wrong tier of support.