AI workforce skills: 85.7% use AI, only 36.7% have been trained

The Singapore figure: near-universal use, training at a standstill

Singapore's digital economy report, published in October 2026 by the public authority IMDA, measures a fact that goes to the heart of AI workforce skills: AI use at work reaches 85.7% of the people surveyed, up from 77.6% a year earlier.

The same survey asks how much training they received. 68% say they need to reskill on AI, while 36.7% took a dedicated course in the previous twelve months^[1].

The two numbers describe distinct populations, with limited overlap. More than thirty points separate those who recognise the need from those who got into a classroom. That space is the real perimeter of the problem.

The figure deserves a note on method: this is a survey by the communications authority conducted among Singapore workers and organisations, with the reference period 2026 for individuals and 2025 for firms. The sample is published in aggregate form, and the historical series allows a year-on-year comparison.

The most cited reason points to a company decision

The most useful question in the survey covers the reasons for missing out on training. The most cited answer, 34.9%, points to their own employer: these people report being left out of the pathways because the organisation left them out.

The other barriers tell the same story in different words:

• 34.1% cite lack of time;
• 26.6% report difficulty choosing the right course;
• 17.8% point to cost;
• 15.1% remain unsure whether their boss would allow the classroom hours.

Time, permission and guidance are variables an organisation governs. People's own willingness appears at the bottom of the list, almost invisible. The skills gap, in this survey, comes from a management decision.

The language of resistance to change holds up poorly against this ranking. People are asking for training en masse, and they point to logistical and authorisation obstacles.

The report itself acknowledges it: closing the distance «may require greater employer support and clearer signposting to suitable training opportunities».

What organisations see: use at the surface

Organisations that have adopted AI give their own version. In 2025 they say 71.9% of their people can use AI in some form, up from 66.4% the year before.

The internal distribution matters more than the total. The largest group, 47.6%, uses AI for generic tasks. 19.3% apply it to problems specific to their own trade and 5% get as far as building solutions for complex work.

People describe themselves the same way: about half often check the outputs, while barely a quarter automate tasks or bring AI inside the workflow. Adoption and mastery are two separate measures, and here they diverge.

The difference between 47.6% and 5% says where processes stop. Generic use produces drafts and summaries, while use applied to the trade changes the way the work gets done.

The reported gains make inaction expensive

People who use AI report concrete results. 72.9% say productivity is higher, 68.9% that the quality of work is better, 61.6% that they have more occasions to spend on higher-value activities.

Among those reporting time savings, roughly six people in ten gain up to an hour a day.

These numbers change the business case for training. An hour a day per person, multiplied across a company population, is a quantity a finance director knows how to read. The cost of the 34.9% left out becomes measurable in lost hours.

The comparison between the two sides is instructive: the 72.9% reporting gains sits alongside the 47.6% judged to be stuck on generic tasks. The value already produced comes from the surface of use, and the margin remains almost entirely intact.

The demand for skills comes from outside the tech sector

The local labour market confirms the direction. Tech employment in Singapore rose 3.8%, to 222,200 people in 2025, and the push comes largely from outside the technology sector.

Sectors other than information and communications employ 59% of the country's tech professionals. Their technical headcount grew 5.3%, against 1.8% inside the sector itself. The share of job postings asking for at least one AI skill moved from 14% to 16.7%.

The digital economy reached 144.1 billion Singapore dollars in 2025, 19.3% of GDP, as CNA's reporting^[2] also notes. Adoption is growing among firms too: from 14.5% to 23.4% among SMEs, from 62.5% to 70.4% among large local companies.

The objection: formal training measures little

A reasonable objection comes from heads of people: learning on the job weighs more than the classroom, and the 36.7% measures structured courses alone. The objection holds a share of truth, and the data on depth of use puts it in proportion.

Informal learning produces the 47.6% that stays on generic tasks. It brings people to the threshold, and leaves the next leap to deliberate design.

The perimeter is worth stating: this evidence covers the people surveyed in Singapore, in an economy equipped with one of the most structured public reskilling programmes in the world. An honest reading holds the geography steady, and looks at the mechanism, which travels elsewhere.

An analysis published by Digital CFO Asia^[3] reads the same picture as a brake on the transformation of the local workforce, with training as the main lever.

What organisations that close the distance do

Organisations that close this space act on three levers visible in the data. They turn implicit permission into an explicit invitation, because a person sent by their own boss walks into the classroom.

Then they reduce the load of choosing: short catalogues, pathways named by role, clear signposting towards the right course. The 26.6% reporting difficulty in selection has a navigation problem, solvable with a map.

Finally they protect time in the calendar. Singapore's authority is widening the TechSkills Accelerator programme on this logic: funding and signposting the pathways reduces the two most cited barriers, time and permission. Organisations that convert their own internal talent start here, before looking at the external market.

What changes for those who lead

Leadership readiness remains the bottleneck of adoption, more than people's own willingness. This evidence offers the top team four concrete moves.

• CEO: bring to the board the share of people sent into training, alongside the share that uses AI.
• CHRO: make the invitation to train a tracked management act, by role and by level.
• CFO: assess the investment against the reported hours saved, up to an hour a day per person.
• Talent and remuneration committee: monitor the distance between declared need and training received as a human capital metric.

The move from adoption to mastery is a designed condition, before it is an individual choice. Whoever governs the design decides how many people cross the threshold.

One design question remains for those who lead: how many people, in their own organisation, received a named invitation to train on AI over the last twelve months? The answer comes from the systems, in an afternoon. The distance between that number and the share that uses AI every day measures the quality of the organisational design, before the quality of the people.