Is AI a headcount killer or a skills accelerator? A new McKinsey survey suggests the near-term answer is: both, depending on where you sit? Most companies saw little change in staff numbers over the past year from AI, but 32% expect headcount reductions in the year ahead as deployments move from pilot to production? At the same time, hiring is accelerating for people who can actually make AI work�data scientists, ML engineers, data architects, and MLOps leads?
What�s shrinking, what�s growing
McKinsey�s State of AI in 2025 report�based on 1,993 respondents across 105 countries�finds early declines in customer care, HR, and sales and marketing roles among regular AI users? Meanwhile, demand is rising for technical and product roles that build and scale AI systems?
- Hiring up: AI data scientists, data engineers, ML engineers, software engineers, AI product managers/owners, and data architects?
- Hiring down: customer care, HR, and some sales/marketing functions where automation can absorb routine tasks?
�Even in these early days of adoption, we are seeing changes in the skills demanded,� said Lareina Yee, senior partner and director at McKinsey Global Institute? She adds that success �requires data readiness and MLOps� (short for machine learning operations, the engineering discipline that deploys and maintains models reliably at scale)? Larger companies, McKinsey notes, are roughly twice as likely to hire for roles that integrate, model, and industrialize data?
Cuts aren�t hypothetical anymore
If McKinsey�s global snapshot hints at a shift, UK employers are already bracing: 26% of large private-sector organizations and 20% of public-sector employers expect to reduce staffing due to AI in the next 12 months, according to the Chartered Institute for Personnel and Development? The pain looks uneven�financial services leads with 37% expecting cuts, followed by IT (26%) and legal/accounting/consulting (24%)?
That contrast matters for planning? It suggests a compositional change in work�junior professional and administrative tasks get automated first�while specialized, higher-leverage roles that turn AI into business outcomes become more valuable?
The market�s reality check on AI promises
Investors, too, are forcing discipline? Last week�s $750 billion sell-off across leading AI-exposed stocks (with one chip champion losing more than $350 billion in market value) reflected two tensions: stretched valuations and uncertainty about who captures enduring profit from AI?
Under the hood, ROI is hard? Gartner and McKinsey data, summarized by the Financial Times, indicate 80% of companies using generative AI report no material earnings contribution yet? Average deployments run about $1?9 million upfront, with hidden costs: roughly 25 extra days for staff training per 100-day project and 100�200 days of post-deployment change management? Add a brittle foundation�around 70% of IT budgets go to keeping legacy systems alive, delaying new features by 6�18 months and consuming 40% of developer time in technical debt?
Case study: shareholder patience has limits
Rightmove, the UK property listings group, saw shares plunge nearly 25% after warning that a step-up in AI spending would slow profit growth? Management argues the investment is central to long-term value creation; analysts called it �two steps back to move three steps forward?� The message for executives is clear: AI roadmaps must connect to measurable outcomes�faster conversion, lower churn, higher lifetime value�or face scrutiny?
The new chokepoint: data supply chains
Beyond skills and CapEx, the data supply chain is becoming a strategic bottleneck? Wikipedia�s operator urged AI companies to stop scraping and use its paid Wikimedia Enterprise API with attribution�both to stabilize infrastructure and to support the content pipeline models rely on? Translation: quality data now has a real price tag and rules of the road, and enterprises need compliant data partnerships baked into their AI plans?
What leaders should do now
- Map functions by automation potential: Identify tasks (not jobs) in customer care, HR, and marketing that can be automated, and redeploy talent toward growth or quality?
- Fund data readiness and MLOps first: Clean pipelines, robust governance, and reliable model ops drive returns more than flashy pilots?
- Set ROI gates and time-boxed pilots: Tie investments to specific KPIs�cycle-time reduction, cost-to-serve, or lead-to-close lift�and scale only when thresholds are met?
- Build a scarce-skills bench: Hire or upskill for data engineering, MLOps, and AI product management; partner where capacity is thin?
- Secure lawful, durable data access: Establish paid or licensed sources and attribution practices to reduce legal and operational risk?
The bottom line: AI isn�t eliminating work so much as reorganizing it? The winners won�t just cut�they�ll recompose? The question for 2026 isn�t whether AI changes headcount; it�s whether your organization can convert that change into productivity and profit before capital markets run out of patience?