Singapore firms are buying AI faster than they can redesign work for it

SINGAPORE businesses are investing heavily in AI, and adoption is accelerating. The more interesting question now is whether those investments are changing how work gets done and delivering value across the enterprise.

Singapore organisations are already allocating 15.4 per cent of their IT budgets to AI, with spending up 108 per cent year-on-year. That share is projected to reach 20 per cent by 2027, according to a recent report by business automation firm ServiceNow. 

The adoption of AI agents, which are trained to take action autonomously on behalf of humans, has also moved quickly. Just over half – 51 per cent – of Singapore organisations now deploy agentic AI, reflecting the rapid take-up of AI in the Republic.

Importantly, 10 per cent have already built processes in which AI can complete multi-step business tasks from end-to-end, according to the ServiceNow study. While this represents an early stage of enterprise-wide AI transformation, it shows that some organisations are already moving beyond individual use cases and putting AI to work across entire workflows.

Companies may now have AI capable of understanding requests, interpreting information and taking action. Yet, the work surrounding that technology can still pass through the same departments, approvals and systems it did before generative AI arrived. 

What seems to be the big obstacles that are hindering success? Part of the answer can be found in where businesses have invested over the past year. 

Singapore is outpacing the rest of the world in building the foundations needed to scale AI – 23 per cent have replaced legacy technology with modern platforms, compared with 16 per cent globally; 28 per cent have formalised AI testing, auditing and risk processes, versus 20 per cent globally; and 25 per cent are integrating workflows across functions, compared with 16 per cent globally.

That helps explain Singapore’s rebound in AI maturity from a score of 34 (out of 100) in 2025 to 53 this year, according to the survey. Businesses are becoming better equipped to deploy AI but that is not the same as redesigning work around it.

AI has made some people faster but it can do more

To understand where businesses go from here, it is worth looking at how they have been using those stronger foundations. So far, this has been for improving individual tasks, such as summarising documents, drafting responses or answering employees’ questions.

While these are good, they are what ServiceNow describes as the “shallowest stage” of using AI to assist individuals instead of transforming how a business works. Prime Minister Lawrence Wong made a similar point at this year’s May Day Rally, observing that AI agents “do not just answer your questions” but “can plan and execute complex tasks from start to finish, all on their own” – a shift beyond simply assisting individuals. 

A customer service officer using AI to summarise a case faster is good, but the underlying process may still be the same. He is one of many people involved in a process that AI has partially accelerated. 

For example, for a return request submitted by a customer, the process would involve operations, finance, and potentially other departments and employees, which can all be improved should AI be deployed across the board. 

The first phase of enterprise AI made individual tasks faster. The next phase has to make the entire end-to-end process better.

Enter autonomous AI agents

Businesses will have to enable AI to work alongside humans and make decisions reliably over time, through learning from humans. In the customer service example, this means AI agents can be acting autonomously at each turn to quickly process a customer request for a return.

Instead of waiting for a human operator to approve each part of the process, a team of well-trained and trusted AI agents can work with one another to work through the process without humans being the bottleneck at each stage.

As processes are improved, AI can also enable businesses to further streamline how their operations are run. To truly get the ROI for AI, they should be able to glean insights from the AI on how to improve over time. 

Says CK Tan, ServiceNow’s Asia-Pacific and Japan innovation officer: “Singapore is entering an important next phase of AI adoption. 

“At the same time, Singapore has been putting practical guidance around the responsible deployment of these more autonomous systems, including the Model AI Governance Framework for Agentic AI, which will need to evolve as agents take on more autonomous, higher-stakes work.” 

For businesses, the opportunity now is to bring those pieces together. He adds: “Adoption needs to be supported by the right data, systems, workflows and governance, so AI can take on more complex work while organisations retain clear accountability and human oversight.

“The question is increasingly not whether businesses have AI, but whether they are ready to make it work across the enterprise.”

Data needs to flow

The next challenge is giving AI agents the context they need to act effectively. Notably, in Singapore, 68 per cent of executives cite inadequate data accuracy and access as their top AI challenge.

An AI agent might understand what needs to be done perfectly well, but if the relevant customer, transaction or employee information sits in several disconnected systems, its ability to act is limited. 

This is where human intelligence is important. In particular, humans compensate remarkably well for organisational friction. An experienced employee knows to ask: “Don’t use that figure; ask Sarah for the latest spreadsheet.” 

That tacit knowledge helps organisations function. Before AI can perform the same workflow autonomously, companies need to make the rules, ownership, information and exceptions explicit.

Not simply automating everything

Where AI works well is to assist and act on behalf of humans. It can gather information, summarise and recommend, as well as complete routine steps within defined rules. Humans can intervene to manage ambiguity, exceptions, sensitive decisions and accountability.

The significant organisational challenge is therefore not simply training people to use AI. It is redesigning roles around the changed division of labour between people and machines.

In ServiceNow’s study, organisations that are doing AI right reflect this. Among these “pacesetters”, 57 per cent provide tailored, ongoing AI training and upskilling, compared with just 4 per cent of other organisations. 

And there is commercial evidence behind that difference: Globally, these pacesetters report 160 per cent ROI on AI investments today, alongside substantially higher productivity and propensity to create new services and revenue channels.

Control at machine speed, orchestration at scale

Just as important is governance, which provides visibility into what agents are doing, what they can access, and the ability to revoke that access quickly if something goes wrong. This has to be built in from the start, rather than bolted on after agents are already running in production. 

That is increasingly described as the AI “kill switch”, though it is not so much a single button that turns AI off, but the ability to isolate the one agent that has gone wrong, among many, without disrupting the rest of the business. 

In practice, a large enterprise will not be managing a single AI but a sprawling mix of agents spread across models, clouds, applications and devices, handling everything from payroll and access provisioning to customer queries and security incidents. If one misbehaves, the challenge is pinpointing and containing that specific agent without bringing the wider operation to a halt.

A kill switch is only as good as the controls underpinning it, and those controls are ultimately a matter of identity and access management working at machine speed: knowing which agents are active, what they are permitted to touch, which systems and data they can reach, and being able to shut off that access the moment something goes wrong.

This means treating every agent the way a company would treat any employee with system access: knowing who it is, what it is allowed to do, and cutting off that access the moment it steps outside its boundaries. 

It is only with that kind of control in place that businesses can confidently orchestrate the broader flow of work across people, systems and AI, so that investment in AI translates into productivity gains and new business opportunities, rather than simply making individual tasks faster.

Says ServiceNow’s Tan: “The bigger opportunity is to ask what an organisation could look like if its workflows were designed for an AI-enabled environment from the start,” he notes.

“That means bringing people, AI, data and systems together around the outcomes the business needs – and moving beyond simply using AI to make existing work faster,” he adds.

None of this is possible without visibility and control: businesses cannot govern agents they cannot see, and cannot act on risks they cannot reach. As AI agents take on more autonomous work, the systems that oversee them will need to keep pace to give enterprises the oversight and the authority to intervene the moment something goes wrong.

Find out more about the 2026 Enterprise AI Maturity Index and what it reveals about the next phase of enterprise AI transformation in Singapore.