AI will change only part of your job. That is exactly why it matters: Opinion

Much of the public conversation about artificial intelligence at work swings between two extremes. On one end, AI is seen as a force that will take over entire jobs and displace workers at scale. On the other, it is dismissed as having limited relevance for many roles, especially those involving care, service, or physical work.

The reality sits somewhere in between. That middle ground is where the real implications lie.

Goldman Sachs economists estimate that AI could automate tasks accounting for about 25 per cent of work hours. This reflects the share of activities within jobs that may be exposed to AI – not a prediction that 25 per cent of jobs will disappear.

At first glance, that figure may seem reassuring. If most of the job remains intact, it may suggest that the disruption is manageable. However, changing even a quarter of what people do at work should not be seen as marginal. In fact, it can reshape how jobs are structured, how performance is evaluated, and which skills matter.

At the same time, the figure also signals a limit. AI should not be pushed into every part of a job. In many cases, the benefits diminish beyond tasks where AI has a clear advantage. This is the central tension: AI is neither a complete substitute for human work nor a trivial add-on. It is a selective force that changes some parts of work fundamentally while leaving others largely intact.

Therefore, the real question is not “Will AI take my job?” but “Which parts of my job will AI change, and how should my job evolve as a result?”

AI changes work. Quietly

This question is no longer theoretical. AI is already changing work, often before organisations fully recognise it.

When I first started using AI in my own research, I expected to save time. What I did not expect was that it would change what I spent my time thinking about. The parts of research that AI helped with most were not the writing – that remained mine – but the work that surrounds it: searching for literature, cross-checking references, reviewing material for consistency.

With that layer handled faster, I found myself spending more time on the questions that actually mattered, and examining whether the framework held, whether the evidence supported the claim and whether I was saying something worth saying at all. The tool did not make me a better writer. It made me a more demanding thinker by clearing the surrounding work out of the way.

Many professionals are similarly experimenting with AI tools on their own. A communications executive drafts a first version of a speech. A finance analyst asks AI to explain anomalies in a spreadsheet. A manager uses it to prepare for a difficult conversation by testing different ways of phrasing feedback. These gains are real, but often invisible to the organisation.

This creates what might be described as “shadow productivity”: efficiency gains generated informally by workers using AI, but not fully captured in workflows, performance systems, or shared learning.

The common assumption is that workers need to catch up. In my observation, the opposite is closer to the truth. Many workers are already adapting – quietly, informally, on their own initiative – finding ways to integrate AI into their daily tasks without waiting for institutional direction.

The real lag is organisational. Firms are holding AI strategy sessions and issuing policy statements, but the jobs themselves remain largely unchanged. The performance metrics are the same. The role definitions are the same. The workflows are the same. What is missing is not worker readiness. It is employer willingness to redesign work in response to what their people are already doing.

This helps explain the growing concern in Singapore about “jobless growth” – where productivity improves, but does not translate into better jobs or broader opportunities. When AI-driven gains remain at the level of individual job tasks, they may not lead to redesigned roles, stronger career pathways, or higher-quality work.

Job redesign revisited

The alternative is not simply more AI use, but better job design. In many offices, AI is already shifting how work unfolds.

In software development, tools like GitHub Copilot can generate code suggestions in real time. Developers report not only writing code faster, but also spending more time reviewing, testing, and integrating what is generated. The job shifts from writing every line manually to overseeing and validating outputs.

In customer service, AI chatbots now handle routine queries such as account information or basic troubleshooting. Human agents are increasingly focused on complex or emotionally sensitive cases. This changes not just workload, but skill requirements – placing more emphasis on empathy, judgment, and problem-solving.

These are not hypothetical shifts, but examples of how tasks within jobs are already being reorganised. These shifts happen when organisations respond deliberately instead of leaving change to individual workers alone.

A marketing role might move from producing content to testing campaigns and analysing performance data. An analyst role might shift from preparing slides to interpreting insights and advising decisions. A manager’s role might expand from supervising tasks to validating AI-supported outputs and guiding team judgment.

Workflow redesign often follows. If AI produces a first draft faster, downstream processes must adapt. Who checks the output? Who is accountable for errors? Who signs off on decisions?

Without these changes, AI simply accelerates one part of the process while leaving bottlenecks – and risks – elsewhere.

The risk of pushing AI too far

Early AI use often delivers quick wins. Drafting, summarising, and basic analysis are natural starting points. But problems arise when organisations extend AI into tasks that require deep context or high accountability without redesigning the surrounding work.

Consider performance management. AI can help organise feedback or improve clarity in appraisal writing. But asking AI to decide who should be promoted or penalised raises concerns about fairness, context, and managerial responsibility.

Or consider financial analysis. AI can summarise reports and highlight anomalies. But relying on it to make investment decisions without human oversight introduces obvious risks.

This is how diminishing returns appear. AI works best when applied selectively and embedded in redesigned processes. It disappoints when treated as a universal solution.

The question is not how much AI can be applied, but where it should be applied – and what needs to change around it.

Readiness, not replacement

If AI changes part of work rather than replacing it entirely, the central issue becomes readiness.

For workers, this means learning not just to use AI tools, but to work with them. In many roles, AI is becoming a collaborator – suggesting options, generating drafts, or simulating scenarios. Workers need to develop the judgment to evaluate these outputs, refine them, and take responsibility for final decisions. This is a different kind of skill from traditional task execution. It requires critical thinking, oversight, and the ability to integrate AI into one’s workflow effectively.

For employers, the challenge is organisational. Firms need to translate task-level changes into redesigned jobs and workflows. This includes updating performance metrics, redefining roles, and equipping managers to lead AI-augmented teams. If this step is missed, productivity gains may remain fragmented, unevenly distributed, and difficult to scale. The role of government should thus be to enable both skills development and organisational change. Training matters, but firms also need support to redesign work and embed learning into daily operations.

Major consequences

The pace of AI change can feel contradictory. At the individual level, adoption is fast. Workers are already experimenting with AI in daily tasks. At the organisational level, change is slower as redesigning jobs, integrating systems, and updating performance frameworks takes time. This is why we may overestimate short-term disruption but underestimate long-term change. The 25 per cent figure captures this duality. It is both a limit and a lever. AI will not replace most jobs outright. But it will change enough of them to require meaningful redesign.

The question is not whether AI raises productivity – it already does. The question is whether those gains are translated into better jobs, stronger roles, and shared outcomes. The risk is not only job loss. It is poor adaptation.

AI will change only part of most jobs. But that part may be important enough to reshape the whole experience of work. That responsibility sits less with workers than most assume. It sits with the organisations that employ them.

Kang Yang Trevor Yu is associate professor at Nanyang Business School and co-director of the NTU Centre for Research and Development in Learning (CRADLE), Nanyang Technological University.

Source: The Straits Times © SPH Media Limited. Permission required for reproduction.

ADV: Newly Published: Singapore Rules of Court: A Practice Guide (2026 Edition) When a deepfake image becomes real harm: Opinion