The bottom of the engineering ladder is thinning out. In engineering, entry-level hiring has fallen to -19% while senior hiring sits at -10%, a 9-percentage-point gap, according to a 2026 GOV.UK snapshot of entry-level hiring. The same squeeze shows up across tech more broadly: the UK tech sector cut graduate jobs by 46 percent in the past year, with a further 53 percent drop projected, according to figures from the Institute of Student Employers reported by The Register.
The timing is not a coincidence. Adzuna has seen a 30% drop in UK entry-level job postings since ChatGPT's launch, with graduates facing the toughest job market since 2018, as techUK reports. As agents start to clear the routine coding queue, the work that remains leans harder on human judgement. That reshapes who teams need to hire, and what they test for.
Demand is concentrating on the specialist and the senior. Demand for AI skills increased nearly 200 percent in a year, with London accounting for 80 percent of AI-related job postings and nearly two-thirds of all technology vacancies in the UK, according to Accenture data reported by The Register. The pull is toward people who can direct and check machine output, not race it.
That demand is broad and durable. Against a subdued 2025 economic backdrop, businesses took a cautious approach to hiring but there were demand hotspots for specialist tech skills including AI, data, enterprise applications and cyber security, as Computer Weekly's tech recruitment outlook sets out. Some roles are already stretched thin: IT business analysts, architects and systems designers, a group of 193k workers, are among the occupations in critical demand, with 3 out of 5 indicators in critical demand in 2025, according to GOV.UK's occupations in demand data.
And the trend has room to run. For the total of AI-related occupations, the net change ranges from approximately 696,000 jobs in the Automation Scenario to 1,072,000 in the Technological Opportunities Scenario, with replacement demand rising from 3,077,000 to 3,193,000 jobs across these scenarios, according to GOV.UK's AI skills labour market projections. The common thread across these numbers is clear: value moves to whoever can judge and correct what an agent produces.
Start by testing the skill you actually need. Instead of a blank-page coding challenge, give candidates agent-generated code with a subtle bug or a weak design choice, and ask them to review it. You learn far more from how someone critiques a pull request than from how fast they type.
Look for three things. First, judgement: can they tell a stylistic nitpick from a real risk to the system. Second, system thinking: do they trace how one change ripples through the wider codebase, security and data. Third, clear reasoning: can they explain why a line is wrong in a way a junior could learn from. These are the muscles that routine automation does not build, so interview for them directly.
Then rethink the junior pipeline. If entry-level routes are narrowing, invest in structured review mentoring so newer engineers grow the critical eye that agents cannot replace. A team that reviews well ships safely, whoever or whatever wrote the first draft.
When you know the skills you need, we help you find the person. We search 15 million candidates to surface engineers who can direct and review agent output, not just write code, and we manage the process end to end. Our recruitment agent manages recruitment end to end for 8% on a successful hire, with no monthly fee and no upfront cost. Tell Reed.ai what great looks like on your team, and we will shortlist against it.
