The AI talent shortage is a portfolio problem, not one impossible requisition
Randstad reports rising demand for AI-agent, governance and human coordination skills across 11 markets. Workforce leaders should separate scarce tasks across build, buy and borrow routes instead of packing them into one role.
What happened
Randstad published a global technology-talent report on 29 October based on 1,100 senior decision-makers, more than 20 million job postings and roughly 100 million candidate profiles.
Why it matters
The findings point to simultaneous shortages in technical orchestration, risk governance and communication, but a vendor survey cannot tell an employer which mix of hiring, internal development or flexible talent will work in its own operating model.
A broad signal, not a local hiring plan
Randstad’s 2026 technology-talent report combines interviews with 1,100 senior business and technology decision-makers across 11 markets with analysis of more than 20 million job postings and about 100 million candidate profiles. The company says 86% of surveyed organisations are expanding technology teams, while demand for AI-agent skills rose 1,300% in its posting analysis. It also reports that demand for risk-governance skills rose 108%, compared with 69% growth in supply.
The report frames the response as buying, building and borrowing talent. Sixty-three percent of surveyed organisations identify flexible talent as their first-choice route for closing AI skill gaps, and 61% say communication, critical thinking and stakeholder-management capabilities are harder to find than technical skills alone. Those numbers support a portfolio question: which capabilities must be resident, which can be developed, and which can be accessed temporarily?
They do not prove that 86% of all employers will add technology headcount, that flexible staffing causes better AI outcomes, or that every market faces the same shortage. Respondents were senior decision-makers, with 100 interviews per market. Randstad is also a staffing provider with a commercial interest in flexible-talent solutions. Its posting and profile data provide scale, but the public summary does not expose every classifier, weighting or country-level denominator.
Check the signal against observed labour data
Revelio Labs’ September AI Labor Market Tracker offers a useful counterweight. It reports that 7.2% of workers in its data have at least one AI skill, little changed from 7.1% in July, and that the number of new AI-adopting firms per month declined 17% from July. Its methodology relies on weighted public profiles and deduplicated job postings, with nowcasting for reporting delays. Revelio explicitly notes that profiles overrepresent white-collar workers and that postings measure intentions rather than realised hiring.
The two sources do not contradict each other automatically. A strong demand signal for selected skills can coexist with slower growth in the number of new adopter firms. One measures a supplier’s global employer survey and market corpus; the other tracks US labour indicators through a different model. The practical lesson is to avoid turning either headline into a single requisition overloaded with architecture, model operations, risk, change management and executive communication.
Build a capability portfolio
Start from the work. List the decisions and tasks needed to deploy an AI use case: workflow design, evaluation, data stewardship, security testing, integration, human review, adoption support and incident handling. For each task, record frequency, consequence of failure, required organisational context and how quickly the capability must be available.
Use those attributes to choose a route. Build internally when the capability is recurrent, context-heavy and important to accountability. Buy through hiring when the work is durable and the market can supply a coherent role. Borrow through a contractor or specialist when demand is episodic, a transfer plan is explicit and ownership remains internal. A flexible worker should not become the only holder of production authority or risk knowledge.
Measure every route with the same evidence: time to independent performance, quality on representative tasks, retention of critical knowledge, supervision load and total cost. Track who can challenge an AI output and who owns the decision after the specialist leaves. A portfolio is working when it reduces uncovered tasks, not when it merely fills seats.
This extends the training-format caution in the JRC AI skills analysis: course supply and hiring signals matter only when they map to actual work and transfer. Keep market evidence as a planning input, not a causal forecast.
Decision rule
Do not open a single “AI transformation lead” role to absorb every shortage named in a global report. Approve the workforce plan only when each critical task has a build, buy or borrow route, a named accountable owner and a transfer test. Revisit the mix quarterly as demand evidence and internal performance change.