The workplace in Singapore is undergoing a quiet revolution. With the National AI Missions accelerating the rollout of agentic AI in priority sectors, organizations are now effectively managing two distinct workforces at once: human talent and AI agents. Singtel‘s recent move to redraw its corporate structure, assigning leaders to oversee blended teams of people and agents, signals this broader shift. Yet acknowledging this hybrid workforce on paper is merely step one. The true operational challenges lie beneath the surface, presenting risks that many boards are not yet ready to confront.
The agentic workforce
It helps to be precise about what an AI agent is. A chatbot answers a question; an agent acts on it, approving an invoice, updating a customer record, or moving funds between accounts. The difference is between an assistant who drafts a letter and a colleague who signs and sends it for you.
What makes an agent valuable also makes it risky. Consider cruise control in a car: it holds the speed you set far more steadily than your own foot, and will hold it straight into the back of a stopped vehicle because judging the road was never its job. An AI agent works the same way. It does exactly what it is asked for, at speed, with no instinct for when something has gone wrong.
Closing the accountability gap
This is where the real challenge begins, and it has less to do with organizational charts than with accountability. When an agent makes a costly error, paying the wrong supplier or breaching a rule no one has set, the agent is not held responsible. The manager who deployed it is. Yet in most organizations, that manager cannot reliably see what the agent actually did, and being accountable for work you cannot observe is a difficult position for any leader.
There is a further risk. Left unwatched, little stops an agent drifting beyond the job it was given: slipping into other processes, reaching into data it was never meant to touch, and acting on what it finds. That is how a useful automation can quietly become a compliance problem. The safeguard is being able to see what each agent is doing, which is what keeps it in its lane.
Singapore’s regulators have put this front and center. IMDA’s Model AI Governance Framework for Agentic AI makes keeping humans “meaningfully accountable” one of its four pillars, warning that an agent’s autonomy can blur lines of responsibility once tied to fixed workflows. The means to observe an agent, review its work, and answer for it has to be designed in from the start, alongside the trust, security, and governance autonomous systems demand.
Governing the invisible workforce
A second, quieter risk is worth naming. Because agents are cheap and simple to create, teams will build their own, each solving a local problem, until the business has an uncontrolled spread of tools with no central oversight, what we have come to call AI sprawl. The equivalent was shadow IT a decade ago, but agents have now raised the stakes because they can act on their own.
The answer is not to ban them, but to bring them into a system where they can be seen and governed. Singapore’s public sector offers a model. As it prepares to put AI agents in the hands of some 150,000 public officers, GovTech is building a registry that tracks who owns each agent and what it does before its use becomes widespread. It is, in effect, accountability infrastructure put in place ahead of scale: you cannot manage what you cannot see.
Structuring for sustainable scaling
None of this argues against agents. The productivity case is real, and those who hesitate will fall behind. What it should change is what leaders must prioritize. The common question is how many agents a business can deploy, and how fast. The better one is whether it can say, at any moment, what every agent is doing, and name the person accountable for it.
Most inefficiency, and most risk, lives in transitions between agents, robots, and people. Addressing these calls for orchestration. Consider an airport: no one lets each pilot pick a runway and hope for the best. A control tower holds a complete view of every aircraft, the authority to set limits, and a record of every movement, so someone can always account for what happened. A workforce of agents needs the same: a single environment where people, robots, and agents work together; each agent’s limits are set, every action is logged, and humans stay in the loop where it matters.
The sequence matters, as the orchestration layer needs to be in place before agents are scaled, not after a regulator or a customer starts asking questions. Firms that rush the agents and defer the oversight will spend years managing avoidable problems. Those that build it first can grow their digital workforce with confidence.
Building a resilient hybrid workforce requires a clear division of labor. In this new model, AI agents think, robots execute, and people lead. AI agents can take on an almost limitless range of tasks, but they can never take responsibility for the outcomes. Accountability cannot be coded; it remains with human leaders. Defining exactly where that accountability sits must be the first step in deployment, not an afterthought.
Karl Crowther is Area Vice President, Middle East & Africa and South Asia, at UiPath.
Editor’s note: An earlier version of this contributed article appeared in The Edge Singapore. This version has been lightly edited for TNGlobal style, clarity and length. The views and arguments expressed remain those of the author.
Share your perspective: TNGlobal welcomes contributed insights and expert commentary from across Asia’s technology and innovation ecosystem. Submit a contribution for editorial consideration, or explore more conversations in our TNGlobal INSIDER and TNGlobal Q&A and Interviews archive.
Featured image: Igor Omilaev on Unsplash
From footfall to foresight: how AI can turn fragmented retail signals into better decisions