SINGAPORE – The rise of artificial intelligence (AI) is kicking off a race among companies to quickly adopt and incorporate the technology. This is spawning new forms of workplace anxiety like “tokenmaxxing”, the act of deploying AI in pointless tasks to comply with mandates to increase AI usage. At the National University of Singapore (NUS), most of its 12,000 or so employees have been receiving a steady stream of e-mails encouraging them to use AI since the start of the academic year in August 2025.
“AI’s meteoric rise has unlocked a wealth of exciting possibilities,” reads an e-mail from NUS president Tan Eng Chye to all staff in May. “Beyond responding with agility, NUS must actively lead and define this transformation. “To ensure this collective endeavour, AI strategy and deployment will be embedded as a leadership KPI for senior management, deans and heads, with the shared commitment extending to every team across the university.” Eight NUS employees, speaking to The Straits Times on the condition of anonymity, say that what
is happening feels at times like the university is putting AI adoption ahead of a clear vision of how AI should improve working life. They note how staff members are using AI to transcribe meeting minutes, write e-mails or create AI-generated art for university events. At the same time, they have also noticed an attendant uptick in AI-related errors that create more clean-up work than if AI had not been introduced. Meanwhile, attempts at incorporating AI into everyday tasks like filtering out postgraduate candidates in
admissions processes have caused glitches that resulted in more manual work, says an employee at the Faculty of Arts and Social Sciences. “The reality is that AI is not advanced enough to help us do anything that’s worth doing in our office,” the employee adds. “We have to expend double the effort to fix it.” In response to queries from ST, NUS deputy vice-provost Bernard Tan says that AI has not been made an official leadership KPI as yet, and that annual performance reviews will
measure quality of work, regardless of whether AI was used. He points to examples of the technology in action. These include AI tools to help staff members pen speeches while maintaining their personal writing style, process large volumes of student feedback, and create interactive simulations for medical, law and social work students. AI for AI’s sake? Currently, the most visible sign of AI in the university is the use of chatbots to replace functions previously performed by humans. The “Servicehub” form used by staff for
IT requests – such as adding new hires onto an e-mail distribution list – has been replaced by an ASK chatbot, while 20 AI agents have been developed to respond to queries like procurement and leave of absence. Part of the issue with deploying chatbots to perform work previously done by humans is that AI responses are prone to errors. Hence, chatbots frequently include a disclaimer calling on users to reach out to the relevant department, which means users often circumvent the chatbots entirely, says
a student-support staff member. In practice, as service requests are ultimately managed and fulfilled by humans, these chatbots present the illusion of work powered by AI, but without clear efficiency gains. Chatbots also create a new form of work when they go off-script, requiring workers to test, refine and add new disclaimers. In other e-mails seen by ST, administrative staff were asked by higher-ups to suggest ideas on how to incorporate AI in the workplace. Teaching staff were also asked to report their level of
AI use in faculty-wide surveys. Some staff have been selected for mandatory AI initiatives like a 12-week training course – which has been overhauled into a shorter course based on staff feedback – and cross-faculty projects where they were given the open-ended task of identifying problems that can be addressed with AI. Not all employees are detractors of these measures, however. “I’m glad the university has finally taken an active stance and signalled that it takes this seriously,” says one staff member who works in
marketing to ST. “While NUS might not be ahead of the curve globally, I’m glad we have at least started.” Planning for the AI future Among local universities, NUS teaches the largest proportion of Singapore’s undergraduate population, with nearly 30,000 undergraduates enrolled across all faculties. Those teaching the next generation of Singapore’s workers say that the aggressive roll-out of AI in their workplace runs the risk of clashing with their ability to critically discuss the technology. In the portal Canvas, used by students to receive
and submit assignments, a standard notice that appears above all take-home assignments states that “use of AI tools is generally permitted for take-home assignments, provided that you clearly acknowledge their use in your submission”. “Faculty are rightly concerned about unreflective use of AI by students in take-home assignments,” says one faculty member. “We used to be able to ban this entirely and penalise it when caught. New policies block penalties of this kind though, provided that AI use is declared, which is frustrating for those
of us who care about incentivising actual student learning and growth.” A member of the teaching faculty at the Faculty of Arts and Social Sciences says these disclosures can often look like a student stating he or she used ChatGPT to generate an outline, Claude to fill in the details and another tool to check for grammar and citations, while claiming all the ideas produced were his or her own. Such AI use has a human cost, says the lecturer. AI use has also become
so normalised among students that even when asked questions like how their internships went, some turn to ChatGPT to produce a suggested response. Conversely, some staff turn to AI to generate lesson plans, grade papers and generate stock responses to student e-mails. In response to queries about this, NUS deputy vice-provost Tan says that surveys asking staff about their AI use are intended to map the current landscape. “In a future workplace, AI will be everywhere,” he says, noting that the university aims to prepare
students for a future when knowledge is not enough and workers are expected to be comfortable with working with AI as a “human in the loop”. He adds that it is ultimately up to students whether they wish to use AI as part of their studies and that policing AI use has become an impossible task. “You can choose not to use AI. Nobody will force you. But the process of completing the assignment will be very onerous,” he says. “But if you rely on
AI for everything, the kinds of solutions you get are sub-optimal. The best outcomes are the ones where you co-create solutions with AI.” For the university, the anxieties surrounding its AI roll-out reflect the enormous stakes of preparing students for the workforce of the future, a professor tells ST. “Everyone is anxious. NUS is anxious about seeming too AI-focused and triggering job anxiety, but also not being AI-focused enough,” she says. “Our conversations are: How will this AI integration support, not reduce, the jobs available
for our students? “Thinking about the jobs in an uncertain AI future and planning backwards from that is nerve-racking.” ‘Tokenmaxxing’ to meet AI KPIs These worries and scenarios playing out at the university mirror those at many large corporations worldwide wrestling with AI’s advancing footprint. At global firms like KPMG, Meta, and Amazon, workers have reportedly engaged in “tokenmaxxing” to meet top-down mandates to improve AI usage statistics. Tokens are the units of data that AI models process. In 2025, financial services firm KPMG’s global
AI workforce lead said in an interview with Bloomberg that staff would be rated on their AI use in annual performance reviews. However, employees speaking to media outlet Business Insider in May said these internal dashboards were easily gamed. “You can just run a prompt. That would be your AI usage for the day,” one US-based KPMG employee told Business Insider, noting that prompts can be automated on weekends to artificially boost one’s usage metrics. For context, AI spending varies widely, depending on the complexity
of the tasks involved. In Singapore, the public sector’s roll-out of an in-house chatbot by Open Government Products (OGP) had an estimated cost of $6.28 per month per active user in the fourth quarter of 2025. However, this reflects the lower end of AI expenditure, as OGP reports that the over 80,000 monthly active users use the tool for an average of 12 conversations a month. In contrast, “tokenmaxxing” from Meta’s 85,000 workers worldwide trying to climb the company’s internal AI leaderboard meant burning through
60.2 trillion AI tokens in 30 days, according to a report by US-based media outlet The Information in April. Charges for OpenAI and Claude’s latest models come in at between US$5 and US$30 per million tokens, though a company purchasing at scale translates to discounts. Incentivising AI adoption without a clearly defined purpose has already led to high-profile U-turns. A few days after The Information’s report, Meta scrapped its internal AI leaderboard. In May, US tech giant Amazon shut down its internal AI leaderboard because
“tokenmaxxing” was inflating expenditures due to pointless tasks that workers were asking AI to do to improve their rankings. “Please don’t use AI just for the sake of using AI,” an Amazon leader beseeched staff members, according to the Financial Times. Speaking to ST, Sima Sadaat, Singapore country manager of tech training school General Assembly, says: “AI-usage KPIs exist, in large part, because organisations genuinely don’t yet know what AI transformation looks like for them.” Unlike past technology roll-outs, where a company could mandate a
specific tool or process and measure adoption clearly, the value in using AI emerges only when individual employees discover, in a bottom-up way, how AI might enhance their roles. This makes it a far more difficult phenomenon to measure, she says. In the end, this means “companies default to what they can measure, which is usage”, she adds. Part of the discomfort with these KPIs is not only the mismatch between the KPIs and what they actually seek to measure (AI-driven improvements to productivity), she
notes, but also employees’ apprehension towards enthusiastically adopting a tool that might eventually make their roles redundant. Lower-value human capital Increasingly, human labour is framed in terms of whether it is more cost-effective than deploying AI. In May, Bill Winters, chief executive of UK-headquartered bank Standard Chartered, stated that the company would be eliminating nearly 8,000 jobs over the next four years in a bid to replace “lower-value human capital” with artificial intelligence. He adds: “We don’t have job losses, but we do have job
role reductions in favour of the machines, and that will accelerate as we go forward into AI.” A public outcry led to Winters issuing an apology three days after. That same month, US-based tech firm Cloudflare’s chief executive Matthew Prince penned an op-ed for the Wall Street Journal titled “How I chose which Cloudflare employees to replace with AI”, after the firm laid off 20 per cent of staff despite record revenue growth. “AI isn’t coming for builders or sellers, but it is coming for
measurers,” he wrote. In the piece, he opined that AI being “tireless, efficient and available” has made many roles in middle management, operations and finance obsolete. Online, comparisons between the cost of human labour and AI have become commonplace, often taking the shape of comparing the cost of outsourced labour or, say, hiring a new software engineer against the cost of tokens. Less discussed is the cost of cleaning up AI output and making it presentable, a friction primarily experienced by rank-and-file workers, rather than
the senior leaders driving AI strategy. To Saadat, employee backlash against AI KPIs reflects wider concerns like a lack of trust in the leadership’s AI roadmap and unclear expectations of what “good” AI use looks like in specific roles. Resistance, she notes, is often framed as “employee reluctance”. But in reality, the constraint is “managerial imagination”. “Many organisations struggle to articulate what higher-value work looks like once AI absorbs lower-value tasks,” she adds. “The companies that get this right won’t be the ones that pushed
hardest on adoption metrics. They’ll be the ones that took the time to bring their people along.”
NUS, artificial intelligence, AI KPIs, tokenmaxxing, chatbots, workplace anxiety, education policy, Singapore