AT two o'clock on a Saturday morning in April, I was trying to resuscitate my AI agent. Everything had been running fine until an update notification popped up. I clicked "update" thinking it was harmless.
The system crashed so catastrophically afterward that it felt less like a bug and more like my agent had decided to rebel against my attempt to improve it.
I copied error messages between different apps and switched between AI tools. I restarted the system. I stared at troubleshooting screens until my eyes burned.
Nothing worked. Eventually I gave up, not because I found the answer to save my AI agent, but because I was physically drenched and could not stay awake anymore.
My conscience kept asking why I had clicked the update button while my brain kept wondering what I should try next.
To rub salt in my wound, I had burned through almost all my weekly AI token limit two days after it reset. I had to wait five more days. I spent my tokens arguing with a broken agent while normal people slept peacefully through their weekend.
On another better weekend in May, I used another agent to build a game – a cancer treatment meets arcade plane shooter game. My hope was that this game could draw attention and maybe funnel donations towards cancer research.
An idea that had existed only in my head was becoming real and playable. Each successful instruction made me want to add one more feature, even as my token balance dropped at an alarming speed.
I started repeating a small mantra: "All is well. At least the tokens are being spent on building something meaningful to fulfill my mission, not paying ransom to a software bug."
That is the strange contradiction of working with AI tools. On a good day, it feels like you have gained several new abilities overnight. On a bad day, you could stay awake from 2am till sunrise because you thought one more prompt might fix the problem.
I recently came across a phrase on social media: "The AI token vampire". It does not drink blood. It engulfs tokens, attention and, occasionally, common sense.
The productivity trap
AI tools have assisted with creations that genuinely matter. People who cannot code like me can now build applications. Small businesses can automate routine work. Employees can explore skills that once required years of training.
Every successful prompt delivers a small jolt of satisfaction, a rush of dopamine in the brain.
Every response an AI tool returns invites another prompt, another revision, and another hour of work.
What I did not notice at first was that AI had removed the natural pauses that once forced me to step away: the water-cooler break, the wait for a colleague to reply or the mental reset between tasks. AI tools run relentlessly and do not stop. They keep churning electrons into activity until you hit your token limit or your budget runs out.
I have been pondering about this a lot lately because this is my new normal. While I am waiting for one task to process, in parallel, I open another chat session to enter the next prompt instantly.
The machine never gets tired nor complain. And somewhere in that rhythm, I stop asking whether I should just keep going because I can.
A 2026 survey by WebMD Health Services found that 80% of more than 3,800 full-time American employees regularly use AI at work. The employees who believed AI made them more productive were also 4.5 times more likely to report burnout, especially the middle managers who are caught between leadership expectations and team realities. They reported burnout at more than three times the rate of individual contributors.
This does not prove that AI causes burnout. But it validated what I experienced with the new normal, where productivity gains can arrive with higher expectations, faster cycles and fewer chances to recover.
Closer to home
Malaysians are not immune. Recently, Human Resources Minister Datuk Seri R. Ramanan told Parliament that 16.91% of workers screened under the National Occupational Disease Prevention Programme were likely experiencing work-related psychosocial disorders.
The report flagged digital technology and AI as workplace culture pressure points, alongside workload management and employment conditions.
Most employees are already operating at full capacity. The expectation is that AI will ease some of that burden. But unless workloads and expectations are redesigned simultaneously, every hour saved will simply be filled with more assignments in the name of corporate efficiency.
I remember designing an AI workflow to ease the workload for both myself and my team.
One of the initial growing pains I observed was that the team had to manage their existing responsibilities alongside the new AI-related tasks and training I had introduced. I take responsibility for future-proofing my team, but I must admit that these processes initially increased their cognitive load in the name of progress.
Employees now process more information and make more decisions without feeling any less busy. They must also verify AI-generated content, learn rapidly changing tools and worry that colleagues who use AI more effectively might eventually replace them.
This is not just an employee wellness issue; it is an operational risk. Exhausted people make mistakes. Overwhelmed managers approve flawed AI output. Teams that begin to associate AI with relentless pressure will eventually resist the very technology meant to help them.
Learning discomfort is not burnout
It is impossible to reverse the AI era, so we will all need to adapt and evolve. This means learning unfamiliar systems, rethinking parts of our professional identities and developing new ways of working.
This requires real effort. And yes, some software updates will still slay my peaceful Saturday night. I would like to emphasise again that there is a fine line between growing pains and burnout.
Growing pains say, “This is difficult, but I am developing.”
Burnout says, “I no longer have the energy or control needed to continue safely.”
The first may be part of progress. The second is a psychological warning that something needs to change.
We need to reimagine workplace culture in the age of AI. We must train people not only to use AI but also to question it, verify its output and recognise when it should be switched off (or alternatively allowed to run overnight).
Corporate performance measures should reward sound judgment and quality. Every productivity gain should not automatically result in a higher target simply because a team can absorb more AI-assisted work.
Burning out the people who use AI is a design flaw. It impairs judgment and creates unnecessary risks for the organisation.
AI is not going away, nor should it. Used thoughtfully, it can help people learn, create and compete in ways that were previously out of reach.
I recently listened to a TED Talk in which Andreas Gebhardt shared a memorable quote: “If you want to learn something, you have to go into the risk, because today’s risk is tomorrow’s safety.”
We are all embarking on an exciting journey in the age of AI with opportunities to develop new skills and solve some of humanity’s most pressing problems. The growing pains may be worth the risk because they equip us with the skills we will need tomorrow.
DR LOW LEY HIAN
Head of Business Development
Cancer Research Malaysia
Subang Jaya
and senior fellow
Pacific Research Centre
Kuala Lumpur
