Singapore's AI Productivity Gap Is an Organisational Problem, Not a Model Problem

Singapore's AI conversation has shifted from whether companies should adopt the technology to why so many of them aren't seeing it pay off. The Ministry of Manpower's latest study puts the gap in stark numbers: 71.5% of firms here have yet to adopt AI at all, and of the 28.5% that have started, only 3.8% have worked it into their core business processes. The rest are running pilots at the edges of the organisation — and that, increasingly, is where the productivity dividend goes to die.

Software delivery is the clearest illustration. The Edge Singapore's analysis of the MOM data points to CircleCI's study of 28 million software delivery workflows, which found that AI is accelerating code production without accelerating the rate at which software actually reaches production. Engineering leaders at Agoda and Zendesk have reached the same conclusion from the inside: writing code was never the true bottleneck, and AI has simply made that visible. Faster code does not automatically become faster delivery when the surrounding system of delegation, review and ownership is still built for human-only teams.

That is the emerging constraint — and it is organisational, not technical. Managers who once coordinated people now supervise a system that blends human and AI execution, which raises unfamiliar questions: which tasks should be delegated to AI, what review standards should apply to its output, and how is accountability assigned when a result emerges from a combination of both? The MOM data suggests the workforce itself is adapting faster than management structures: only 6.2% of AI-adopting firms reported reduced headcount, while 18.9% redesigned job roles and 13.9% created new AI-related ones.

The implications for Singapore's broader push are significant. The refreshed National AI Strategy and the newly formed National AI Council are premised on turning AI access into measurable productivity — and that bet now rests on management capability as much as model capability. Teams are already generating more output than their review pipelines can absorb, and more tools will not fix that. The role of the engineering manager is shifting from deciding what cannot be done to directing what should be: prioritising modernisation, retiring technical debt, and channelling the new execution capacity somewhere useful.

Why it matters for Singapore: A national strategy that promises AI-driven productivity gains lives or dies on how companies redesign work, not on how many models they license. MOM's numbers — 70.7% of adopting firms reporting worker productivity improvements, with displacement confined to just 6.2% — suggest the workforce is ready to be re-tasked rather than replaced, which makes delegation and review design the binding constraint. Expect the local conversation to shift over the coming months from AI access to AI management, as firms that captured the easy wins discover that the next layer of gains requires rebuilding how managers manage.