AI training is going to the wrong people

PwC's Global Workforce Hopes and Fears Survey 2026, published on 29 September 2026, asked tens of thousands of workers across dozens of countries about their year. More of them are using AI, and the share who can get training fell to 51%. The largest group in the workforce, with the least scarce skills, is the least likely to be given any, while the small group furthest ahead is the most rewarded and the most likely to leave. Singapore's own labour data, published on 21 September 2026, shows the same shape from the other end: retrenchments up, and laid-off residents taking longer to find work again.

The two headline movements in PwC's survey point in opposite directions. Use of AI at work rose sharply over the past year, and the share using generative AI every day grew even faster. Over the same period, the share of workers saying they could access the learning and development they need fell to 51%. PwC's reading is that adoption has outrun the ability to use it well. Most workers expect their AI use to increase again over the next year, and they report the technology leaving them energised rather than tired by a wide margin. The appetite is there; the support is receding.

PwC sorts the workforce into four groups by how much AI has improved their work and how much demand there is for their skills. At one end are the front-runners, with scarce skills and strong AI capabilities; at the other, the largest group, which has neither scarce skills nor much AI experience and does most of the day-to-day delivery. That group is the least likely to be offered training. It is also the only one more likely than average to see avoiding mistakes rewarded, while the front-runners are far more likely to see using AI well rewarded. The line between them is drawn by what an organisation pays attention to, not by what the technology can do.

The rewards of getting in early show up in confidence as well as output. Workers who use AI daily report more confidence in their job security than infrequent users, are more likely to trust top management and more certain they can learn new skills. The front-runners are far more likely than average to ask for a promotion, and a meaningful share of them say they are very or extremely likely to change employer within the year. PwC's global workforce leader, Peter Brown, put the risk plainly in the report: 'There is a real risk that the global workforce is starting to move at different speeds.' For an employer, the group that has gained most from AI is also the group most likely to take that gain elsewhere.

PwC's survey also records two pressures that have nothing to do with AI. Only a third of workers report having money left at the end of the month, and most say the cost of living has had a moderate or major impact on them at work; more than a quarter say feeling fatigued or burned out limits their productivity. Asked what threatens their job security, workers name economic volatility first and AI taking on more tasks second. A plan that treats AI training as the whole answer is answering the second question on the list.

That ordering matters, because it says the thing workers fear most is not the technology but the economy around it.

There is no local equivalent of PwC's survey, so we will not pretend the global percentages describe Singapore. What the Ministry of Manpower reported on 21 September 2026 is harder-edged and about jobs rather than attitudes. Retrenchments rose again in the second quarter of 2026, concentrated in manufacturing, information and communications, and financial services, and in most cases the reason given was business reorganisation or restructuring. The re-entry figures are the part that speaks to training: the share of retrenched residents who found work within six months fell to 54.9%, and it was degree holders and residents in their fifties who fell furthest. When people lose work and take longer to return, the training they did not get shows up as a slower recovery, not as a survey answer.

That is the version of the training gap that costs money rather than goodwill.

The finding we would act on is the least dramatic one: the people who most need training are the least likely to get it, and the people who get it were already ahead. Most organisations roll AI out to their volunteers, because volunteers make a pilot look successful, and that is how a training gap becomes a permanent one. Sending the same licence and the same hour to the largest group is harder, and it is the only version that moves the average.

We are less convinced by the confidence gap. Feeling secure and trusting management are good outcomes, and they are also what you would expect from people who were handed a tool, a budget line and some attention; PwC notes its groups rest partly on how workers rate their own skills, and the survey cannot establish cause. The hard pressures it records — burnout limiting productivity, nothing left at month end, trust in decline — are not things we would try to train our way out of. For a studio here the practical version is narrow: name the person who owns the result before a tool is bought, send the person who is not the enthusiast to the workshop, and count the rework rather than the number of uses.

The confidence numbers are worth watching over time, but on their own they tell us more about who was given attention than about who was properly equipped.