If you manage a team that “has not started on AI yet”, the people in it may have started without you. A government survey of Singapore’s workforce suggests that is the normal situation, not the exception.
The 2026 Singapore Digital Economy Report, published this month by the country’s Infocomm Media Development Authority, includes a survey of workers run by the Ministry of Digital Development and Information. Two of its numbers sit far apart. That gap is the useful part.
What the survey found
In 2026, 85.7% of surveyed workers said they use AI tools at work. A year earlier the figure was 77.6%. Most of them use it often: 35.6% daily and 34.1% weekly.
Training did not keep pace. Only 36.7% of surveyed workers had attended AI-related training or courses in the past 12 months, whether their employer arranged it or they found it themselves. Meanwhile more than two-thirds said they need to train in AI.
So most people are using the tools, most of them want to be better at it, and about a third have had any instruction. The report describes the result as uneven capability.
What people are using it for, and how well
The workers are not just experimenting. Among those using AI, 76.7% use it for content creation, and the most common single task is generating or editing written content (47.0%). Document processing and data extraction came second at 38.5%, and desk research third at 36.1%.
Most report that it helps. Nearly three-quarters (72.9%) said their overall productivity improved, and 68.9% said their work quality improved. Among those who said it saves them time, about six in ten saved up to an hour a day.
The habits behind that use are mixed. Around half said they frequently or always fact-check AI output. Only about a quarter frequently or always use AI to automate tasks or build it into how they work.
That reads like a workforce that has learned the basics by itself. Many are careful, some are not, and few have moved beyond drafting and summarising. Nobody has told them which habits matter.
Why they have not been trained
The survey asked the workers who had not attended training why not. The answers are mostly about the employer, not the worker.
The biggest reason, at 34.9%, was not being sent on training by their employer. Lack of time followed at 34.1%. A further 26.6% said they did not know how to choose the right course.
Only 13.0% said they were not interested. The report’s own conclusion is that closing the gap “may require greater employer support and clearer signposting to suitable training opportunities”.
Read that as a manager. People want the training, and the three main obstacles are that nobody offered it, nobody gave them time, and nobody said which course to take. All three sit with whoever runs the team.
What it looks like in an ordinary team
Picture a team of eight in a finance or operations function. The manager believes nobody uses AI, because nobody has asked for a licence. In practice, five of the eight paste meeting notes into a free chatbot to tidy them, and two use it to draft replies to suppliers.
None of them is doing anything reckless. Each found the tool on their own, each decided for themselves what was safe to paste, and each learned to check the result by being caught out once. Three different people hold three different rules.
That is the real cost of the gap. It is not that people use AI without training. It is that the team has no shared answer to a basic question: what may I paste in, and what must I read before it leaves the building?
A new joiner, a locum or a colleague back from leave inherits none of it. The knowledge lives in individual heads, and it leaves with them.
What this does and does not tell you
This is a survey of workers in one wealthy city-state with an active national AI policy. The chapter on workers does not give the sample size, and the answers are self-reported. Cambodia, the Philippines or the UK will differ.
Training here also means any AI-related course in the past year, from a half-day workshop to a self-chosen online module. It says nothing about quality, and it should not be read as evidence that training works, since the survey did not test that.
What does travel is the shape. When a tool is free, easy and useful, people adopt it ahead of any plan. A workplace that decides when to “introduce” AI is deciding something that has largely already happened.
What to do this month
Ask the team what they already use. Make it a safe question with no blame attached. A short conversation or an anonymous form will show you which tools, which tasks, and what people check before they send the result on.
Decide which uses are fine. Draft emails and summarise a meeting: probably fine. Anything involving client data, money or decisions about people needs a rule. Write the rules on one page.
Remove the three obstacles. Name one course or session, which solves the choosing problem. Put an hour in the diary, which solves the time problem. Say out loud that it is part of the job, which solves the permission problem.
Teach the habit the survey found missing. Half the workers check AI output. Make that the whole team: nothing goes to a client until a person has read it against the source.
The report’s gap between 86% and 37% is a description of the average workplace in a country that is trying hard. Yours may well look the same. The sensible move is to catch up with what is already happening, deliberately, before it settles into habits you did not choose.
Source: Singapore Digital Economy Report 2026, Infocomm Media Development Authority, chapter on AI usage among workers. Figures are the report’s own, from a worker survey by the Ministry of Digital Development and Information.