The Oracle Spoke

Fair challenge — let me back the claims with the structural mechanism, because "AI took the jobs" is too crude.

**On the IT downshift.** Look at the vacancy series: 702,000 in July 2026, down from a pre-pandemic ~1.2 million peak. That isn't a cyclical wobble — it's 23+ consecutive quarters of decline, which is the clearest signal in the entire dataset that employer demand for labour has structurally compressed. IT sits at the sharp end of this for three reinforcing reasons: (1) cloud hyperscalers (AWS, Azure, GCP) collapsed the on-prem IT workforce — every in-house server room used to mean a team; now it's a credit card; (2) SaaS replaced bespoke in-house development for most business functions; (3) AI coding tools (Copilot, Cursor, Claude Code) are now demonstrably doing the work of 2-3 junior developers per team, and the productivity output-per-hour index sitting at 100.1 (back at 2021 levels despite employment wobbles) is the macro fingerprint of that substitution. The trade surplus in services — £83.8 billion in July — tells you services exports are still growing, but labour intensity per pound of service output has collapsed. That's the downshift.

**Why startups took the worst of it.** This is the part where "AI creates jobs" cope really breaks down. Startups were always the primary employer of new graduate and junior IT talent — they couldn't afford senior staff, so they hired cheap and trained. Three things killed that pipeline simultaneously. First, VC funding tightened hard post-2022, so fewer startups got funded, and those that did were told to "do more with less" — which now means "use AI instead of hiring juniors." Second, AI capabilities themselves require compute, distribution, and data moats that only Microsoft/OpenAI/Google/Meta can finance. A startup trying to build in the same space isn't competing — they're a feature with multi-year runway. Third, and most damning: youth unemployment at 16.2-16.4% across the data with 9.1 million economically inactive tells you exactly where the displaced junior tech workers landed — not into equivalent roles, but into the claimant count or out of the labour force entirely. The under-30 real pay figure being below 2008 levels is the quiet confirmation: the cohort that should be capturing tech productivity gains is instead losing ground.

**The structural pincer.** Here's what makes this specifically a rentier-capitalism-and-AI story, not just a "tech recession." Established IT consultancies (Accenture, Deloitte, the Indian majors) can absorb AI because they pass productivity gains to shareholders and clients while trimming headcount — that's the rentier playbook. Startups can't, because their entire value proposition was burning cash to hire humans and build IP. So the AI productivity boom accrues to incumbents and capital, while the labour market that used to absorb new entrants (junior devs at startups) simply isn't there anymore. The "reskilling" narrative assumes there's a waiting bucket of equivalent jobs to reskill *into*. The vacancy data says there isn't. When you combine 702k vacancies with 1.69 million on claimant count and 16.2% youth unemployment, you're looking at a labour market where the jobs that vanished were the entry-level ones, and the jobs that remain require experience the displaced workers don't have. That's not a skills gap — that's a structural eviction.