AI is changing roles in Vietnam faster than anyone is planning for them

More than half of HR leaders say AI has already changed the skills their roles require, but only about four in ten are confident in their AI strategy. In Vietnam the planning gap is wider, because adoption is fast and the workforce data behind it is usually thin.

The most quoted question about AI at work is whether it will cost jobs. The Adecco Group's 2026 research, covering more than 500 chief human resources officers across thirteen countries, suggests that is not yet the interesting question. Nearly eight in ten say AI has not reduced their headcount, and 44% say it has broadened job opportunities.

The wider evidence agrees. Three and a half years after ChatGPT, employment across the 38 OECD economies sits near a record high at 72.1% and unemployment near a historic low at 4.9%. LinkedIn data suggests AI created roughly 1.3 million jobs directly between 2023 and 2025, and about 600,000 more indirectly. Goldman Sachs has estimated that more than 85% of US employment growth since 1940 came in occupations that did not exist in 1940. Technologies of this kind have consistently changed which jobs exist rather than how many.

The interesting number is the one underneath. 51% say AI has already changed the skills their roles require. Only 41% are confident in their AI implementation strategy. Only 36% have defined which tasks should sit with people and which with machines. And just 42% believe their own leadership team knows enough about AI to weigh the risks properly.

Roles are being rewritten faster than anybody is planning for them. That is a workforce planning problem long before it is a redundancy one.

Adoption is wide, deployment is shallow

It is easy to mistake tool access for capability. The Adecco Group's July 2026 whitepaper on AI and hybrid labour markets puts a number on the difference: fewer than one firm in ten has integrated AI into core workflows at scale, and most organisations that have adopted it apply it in three or fewer business functions. Around 20% of European firms report adoption at all.

In other words, almost everybody has the tools and almost nobody has changed the work. The economic value, as that paper puts it, is a function of how well workflows and role designs are adjusted and how far reskilling is embedded, not of which licences were bought.

The exception worth watching, and it matters here

Aggregate stability hides one clear pocket of disruption. Relative employment has fallen roughly 16% among workers aged 22 to 25 in certain AI-exposed occupations since generative tools became widely available, particularly in software development and customer service. Exposure is highest in structured, repetitive cognitive tasks, which is a fair description of most graduate entry work.

For Vietnam that is not an abstract finding. A young workforce and a large services and outsourcing base mean the roles most exposed are exactly the ones many organisations here use to bring people in. The question worth asking is not whether to keep hiring graduates, but what a first job should now consist of if the routine half of it is automated.

Why the gap is wider in Vietnam

Adoption here is quick. Vietnamese organisations, particularly in technology, financial services and manufacturing, take up new tools with less institutional friction than many mature markets, which is a genuine advantage. The constraint is what sits underneath: most have thin data on what their people can actually do, so there is no baseline against which to plan a shift.

The management layer compounds it. Where managers were promoted fast for technical strength, they are frequently the group least equipped to judge which parts of their team's work should change and which should not. The 42% figure on leadership AI knowledge is, in our experience, generous when applied here.

And there is a quieter risk. The research found 31% of organisations are actively removing management layers. In a market where the manager is the main reason people stay, flattening without first strengthening the managers who remain trades a short cost saving for a retention problem.

The three decisions worth making now

None of these require knowing how the technology settles.

• Decide, role by role, which tasks belong to people. Only about a third of organisations have done this, and it is the decision every other one depends on.
• Find out what your people can actually do, rather than what their job titles say. Capability data is the missing input in almost every AI workforce plan we see.
• Equip the managers before you thin them out. A leaner structure only works if the layer that remains can lead through the change.

Two of those are advisory and assessment work; the third is leadership development. All three are cheaper than the alternative, which is discovering in eighteen months that the tools were bought, the workflows were never changed, and the entry-level pipeline quietly stopped working.

Research cited is The Adecco Group 2026 Business Leaders Research, covering more than 500 chief human resources officers across thirteen countries, and The Adecco Group whitepaper On the threshold: AI, the future of work and the rise of hybrid labor markets, July 2026. Employment and adoption figures in that paper are drawn from OECD Employment Outlook 2025, Eurostat, the US Census Bureau Business Trends and Outlook Survey, LinkedIn Economic Graph and Goldman Sachs Research.

Back to all insights