If you feel your team is behind on AI, the most useful number published this week is 3.8%.
That is the share of Singapore firms that have integrated AI into their core operations, according to the Ministry of Manpower’s 2026 establishment survey. The rest, the report notes, mostly “remained at the planning or pilot stage.”
The figure appears in a report published on 30 September by AMRO, the ASEAN+3 Macroeconomic Research Office. AMRO is a regional body set up by the governments of Southeast Asia, China, Japan and South Korea to monitor their economies, and it has nothing to sell.
Singapore is the hard case, which makes it the useful one
Singapore is not a slow adopter. It has pushed AI across its economy earlier and more deliberately than almost anywhere in the region, and the figures show it.
Adoption among small and medium-sized enterprises more than tripled in a year, from 4.2% in 2023 to 14.5% in 2024, according to IMDA, Singapore’s digital development regulator. Among larger firms it rose from 44.0% to 62.5%.
The Ministry of Manpower survey found 28.5% of firms had begun adopting AI. The rate rises sharply with size, from 23.9% of firms with fewer than 25 staff to 76.4% of those with more than 500.
So a lot of firms have started, and very few have gone further. If that is the position in the region’s most advanced market, it is almost certainly the position in your sector, wherever you are.
What “using AI” mostly means
IMDA’s figures also show what adoption looks like in practice. Among firms already using AI, 84% use off-the-shelf generative AI tools. Fewer than half have adopted anything customised or built for them.
In plain terms, most organisations that say they use AI have staff using a chat assistant. That is a reasonable place to be. It is not the same as having changed how the work gets done.
There is a simple test for which side of that line a team is on. If the chat tool disappeared tomorrow, would any process stop, or would people go back to doing the same work the slow way? If it is the second, you are where more than 96 in every 100 Singapore firms are.
Moving into core operations looks different. A process is designed around the tool, someone is responsible for it, there is a defined check on what it produces, and there is a fallback for when it fails. None of that requires custom software. It requires someone to decide it is worth doing properly.
AMRO is careful about the distinction. Beyond initial adoption, it writes, “few firms have begun using AI to transform their core business processes.” The barriers it lists are not exotic: high implementation costs, limited in-house expertise and concerns about data governance, felt most by smaller firms.
The jobs finding is calmer than the headlines
The report also looked at what happened to work in firms that had adopted AI. Only 6.2% reported reduced headcount. Three times as many, 18.9%, reported redesigning jobs, and 13.9% reported creating AI-related roles.
That fits what most managers see. The change shows up inside jobs, in which tasks a person does and which they check, well before it shows up as a smaller team.
AMRO does flag one group to watch. “Entry-level roles often involve routine tasks that provide graduates with their first pathway into professional occupations,” it notes, and those routine tasks are the ones AI is taking on first. If your team relies on junior roles to train its future seniors, that pipeline deserves thought now rather than later.
Why waiting is not the safe option
Faced with a fast-moving field, it is tempting to wait for the tools to settle. The evidence suggests they will not.
This week Google announced its newest model, Gemini 4 Argon, and is releasing it in stages. It goes first to a set of trusted cyber defenders, then to paying developers and its top subscribers, with everyone else to follow “as soon as possible”. The version on your team’s screens is never the frontier, so waiting for the next one is a permanent state, not a plan.
AMRO’s long-run numbers make the same point for a whole economy. It estimates AI could raise Singapore’s potential output by 1.06% in the long run, with 0.65% realised by 2040 and 0.78% by 2050. Those gains take decades to arrive, and they go to the organisations that move from trying to building.
Three steps from trying to building
List what your team actually uses. Every AI tool, and for each one, mark whether it touches core work or sits beside it as a chat window. Most teams will find the second list is the whole list. That is an honest starting point, not an embarrassment.
Pick one core process, not five pilots. Choose a piece of work that matters, happens often, and has a right answer someone can check. One process redesigned properly teaches more than five experiments that nobody owns.
Name the barrier. AMRO’s three are cost, expertise and data governance. Write down which one is actually stopping your team. It is rarely the one people assume, and each has a different fix.
On the source
AMRO’s analysis draws on two Singapore government surveys, by the Ministry of Manpower and by IMDA. The adoption figures above are theirs, as reported in AMRO’s selected issue paper on AI adoption, and they describe Singapore firms rather than the region as a whole.
The output estimates are AMRO’s own modelling. AMRO describes them as calibrated scenarios rather than forecasts, so treat them as an indication of scale and timing, not a prediction.
Fewer than four in a hundred firms in the region’s most advanced market have built AI into their core work. The boat has not left. Most people are still on the pier, working out where to start.