AI fluency will become baseline skill, but employees shouldn’t treat it as tool

Workers increasingly have to be able to understand artificial intelligence (AI), but not simply as tools they use to execute orders. They also will need to work together as collaborators, while still maintaining oversight of the technology.

AI fluency will become the baseline skill in 2026, much like gaining the ability to understand, speak, and communicate with a language, said Peter Finn, generative AI (GenAI) and cybersecurity instructor at Vertical Institute.

Fluency here will encompass having an accurate assessment of AI and understanding its limitations and risks, because the technology comes with significant limitations, Finn said in a video call with FutureCIO.

AI itself does a good job of masking and overstating its confidence in things, so it may not always inform users when it does not know the answer to certain questions, he said.

This means a certain level of fluency, on the human's part, is needed to understand the limitations and question the AI deeper.

Finn highlighted the importance of “trust calibration”, where there is a correct understanding of when someone should trust the AI and when they should question it more.

Much of the fluency in AI will come through actual experience, which encompasses continued experimentation and applying the technology to as many context and situations as possible, he said.

People expect a standardised playbook on how to use AI in certain situations when, in fact, such guides still are being written as the technology continues to evolve, said Finn, who teaches GenAI and cybersecurity courses at Vertical Institute.

Fluency in prompt engineering and interactions with AI, for instance, were developed as users explored how they should communicate with AI tools. None of these techniques were programmed into the AI itself, he said.

Transforming workforce goes beyond AI

Workforce transformation starts with work design, not software selection, said Jessica Zhang, Asia-Pacific senior vice president at ADP, which provides cloud-based HR management services.

Asked about common pitfalls organisations should watch for as their workforce evolves in an AI era, she noted that organisations too often invest in tools without first redefining job structures and how teams collaborate.

“Sustainable transformation requires more than technology,” Zhang said in an email interview. “Leaders must pair AI deployment with structured mentorship, psychological safety, and clear career pathways.”

“When transformation is treated purely as a training challenge, organisations overlook the deeper behavioural and managerial constraints that hold people back," she said.

In its 2026 HR Trends Guide, ADP said organisations are enhancing their HR capabilities with agentic AI, which creates opportunities for stronger AI governance and HR-IT collaboration.

Zhang noted: “The workplace of 2026 will be intelligent, interconnected, and human centric. Technologies, such as AI, are becoming integral to daily operations and will continue to shape how work is done across Singapore. At the same time, these technologies will fundamentally change how we build skills, design roles, and manage HR processes.”

While agentic AI will reshape workflows more profoundly due to its autonomous capabilities, its adoption ultimately depends on employee understanding and trust, she told FutureCIO.

As it is, ADP’s People at Work 2025 report revealed that 19% of workers in Singapore were unsure how AI would affect their jobs. This uncertainty was highest amongst knowledge workers, at 26%.

Zhang recommends that organisations adopt a practical transition by first introducing AI agents as collaborators.

Initial use cases should focus on low-stakes, common tasks, such as summarising, drafting, and scheduling, so employees can experience clear benefits and build confidence in agentic AI.

“Establishing clear escalation boundaries is also critical,” she added. “Employees who fear replacement experience higher stress and are more likely to consider leaving. Transparent communication on roles, responsibilities, and accountability frameworks can reduce anxiety and strengthen trust.”

She noted that employees who feel supported and fairly assessed adapt more smoothly to AI-enabled workflows.

Singapore's support gap, where lack of support rather than lack of skills, is the biggest barrier in advancement, she said.

“A key takeaway for employers is that AI should elevate employees, not add pressure,” Zhang said. “Employers who combine transparency with strong development support will unlock higher performance.”

She further urged organisations to take a human-centric approach to AI.

“While AI can reduce manual workload, it only strengthens decisions when humans guide and oversee its outcomes,” she said. “Without proper oversight, risks such as errors and bias can undermine productivity.”

AI can undo work, just as it speeds up

Agents have tremendous capabilities to get things done, but also can undo work very quickly, said Finn, who uses AI agents routinely to support his coding work.

When they function well, AI agents can help save significant time. However, they can undo hours, and even days, of work if they pursue a flawed trajectory, he said.

“So maintaining human oversight over AI as a fundamental need hasn’t changed,” he stressed. “Put in the frameworks, put human in or on the loop, where humans have to authorise what the AI recommends or where they monitor and periodically intervene.”

Asked about vibe coding, Finn said his team relies on AI as a coding assistant. However, while it can be a powerful tool, AI has yet to reach a level of reliability where it can be left entirely on its own to generate a piece of software.

“There’s absolutely no way I or my team can build without knowledge of software development, or DevOps, or cybersecurity,” he said.

He explained that several of these AI code assistance tools exhibit poor security practices with regards to what they code, despite being knowledgeable of cybersecurity rules.

“It can flag many vulnerabilities in the code it itself had built,” he said.

The quality of its coding is further questionable if the software has to be integrated with other applications, such as a company’s existing legacy systems or API (application programming interface).

It renders vibe coding impossible in such scenarios, Finn said, adding that the practice is limited where significant systems integration is required.

If you become over-reliant on AI tools, you lose your critical thinking abilities and deskill to a certain point.

Peter Finn

And while AI can improve as it continues to learn and train, it may focus more on some information or context over others, Finn said.

He related how an AI coding tool, after being given instructions, proceeded to build using the wrong software library, while another did not read the relevant software library to build. 

Any respectable software engineer would not build a piece of application without first reading the software library, he said.

While these issues can be addressed with more and better training, Finn noted that AI, at the same time, will be tasked to do more and to carry new tasks.

AI training is never complete, and there will always be a problem that it has not seen before, he said.

He urged the need for users to adopt a “healthy scepticism” with AI.

Need to retain critical thinking

Finn further stressed the need for employees to assume a more critical approach, so they do not blindly accept AI outputs.

This also ensures humans retain their critical thinking, he said.

“If you become over-reliant on AI tools, you lose your critical thinking abilities and deskill to a certain point [where you risk losing cognitive capabilities],” he explained. “Having AI think for you is detrimental to the person doing the offloading, [so] using AI in a more collaborative way keeps the [human] brain engaged.”

Describing it as “human-IT teaming”, he said the principles here can be taught quickly but requires practice, and with AI regarded as a collaborator, rather than a tool.

“You might ask the AI what it thinks at the end of a task or project and it might come up with interesting observations, something that humans may miss,” he said, as he likened communicating with AI to communicating with a colleague.

He noted that this enables humans to remain competitive as they become effective collaborators.

Ultimately, he added, it is about building stronger teams where humans can compensate the AI’s weaknesses, and vice versa.

In his speech last November at the 32^nd APEC Economic Leaders’ Meeting, Singapore’s Prime Minister and Minister for Finance Lawrence Wong underscored the need to build up the capabilities of the country’s workforce and prepare workers for AI.

“With every new wave of technology, some jobs will disappear and new jobs will be created. Historically, the net impact has been positive, meaning more net jobs have been created,” Wong said.

“But there is no economic law that guarantees this will automatically happen,” he said. “Just because it happened in the past doesn’t mean it will also happen in the future. We cannot leave this to the market.”

This meant that governments must invest more in redesigning as well as creating jobs, and reskilling workers, so “we are not caught on the backfoot”, he noted.

Doing so also will ensure the returns from AI investments will accrue not only to organisations, but also to workers and citizens, he said.

Clarity over AI impact key to workforce transformation

Reiterating that uncertainty over how AI would affect their jobs was highest amongst knowledge workers, Zhang noted that this finding challenges the common assumption that highly skilled workers feel more secure during technological shifts.

“While employees recognise AI’s potential, many lack clarity on how their organisations plan to apply it,” she said. “Clear communication, transparency, and regular sentiment checks can reduce speculation and help workers adapt with confidence.”

According to the ADP study, perceptions of AI-driven productivity differed by industry, with professionals in technology, finance, and insurance across more confident that AI would enhance their work.

Optimism also was strongest in roles with deeper technical content, which Zhang said suggested that job design significantly shaped expectations of productivity.

Trust, she added, has become central to employee experience as workers seek clarity on how AI will affect their careers.

They also want more transparency on how decisions are made and whether new technologies can support their long-term development, she said.

Technology alone, though, does not drive competitiveness. “Organisations that combine AI with strong workforce strategies will achieve the greatest impact,” Zhang said.

As AI workloads expand and edge computing becomes integral to inferencing, cost structures will shift towards balancing technology investments with workforce optimisation, according to Lyon Poh, partner and head of corporate transformation at KPMG in Singapore.

“Organisations will need to rationalise ROI (returns of investment) of leveraging human talent versus AI agents to achieve an optimal mix,” Poh said in a previous chat with FutureCIO on AI cost management.

It will ensure productivity gains are matched by sustainable cost efficiencies, he added.

Organisations that integrate AI agents into workforce planning will unlock efficiencies while preserving human oversight, governance, and creativity, he said. The goal should not be cost minimisation, but value maximisation.

Workforce readiness is critical, he noted.

“AI adoption is not just a technology shift; it’s a cultural and organisational transformation,” Poh said. “Reskilling, governance, and change management are essential to embed AI into business processes effectively.”