Coding is dead. Cybersecurity is dead. And as of this writing, IBM’s old Cobol programming language is screwed as well, since AI can manage it soon.
Those are some of the click-bait proclamations you’d find on social media, particularly on LinkedIn and X (formerly Twitter). Posted by AI proponents, they sometimes come with a link to get on the courses and consultancies they sell. This is AI branding in 2026.
With Anthropic’s Claude now so good at coding, some believe that AI can enable anyone to “vibe code” any program for them. AI agents are powerful enough today that they can “red team” cyber defences automatically, without hiring an expensive cybersecurity company to stress-test.
As for Cobol, just Anthropic saying that AI can modernise the decades-old programming language running the mainframes (used in banks) has wiped billions of dollars off IBM’s stocks.
And then there are the Big Tech executives, each proclaiming that AI is going to reorder the economy. Nvidia’s Jensen Huang tells people to be plumbers (not his children, who are working in Nvidia). Microsoft’s AI head, Mustafa Suleyman, says virtually all white collar tasks will be automated in 18 months.
So when the Singapore government unveiled a Budget that pushes hard for AI skills a fortnight ago, you can be forgiven for asking what jobs you will be retraining for. If white collar jobs are gone next year, is it better to learn plumbing instead of going to an AI course?
No less than the prime minister himself is heading Singapore’s AI quest. He’s helming an AI council to make sure citizens do not miss out on the opportunities and suffer the disruption that is already upon us.
Entire industries will get help to accelerate their AI adoption. And premium AI tools will be made available to Singaporeans who undergo selected training courses. The aim: Plug into the new technology on a large scale, quickly.
The approach is the right one, though it also exposes the AI paradox today. AI is so good yet so bad, so ready to unlock a future yet seemingly only good at destroying jobs by scraping up all the human knowledge at great cost to the environment and more.
Much of the hype is scary because it plays to the uncertainty that any disruption brings. Yet, that same hype often does not survive scrutiny once AI is used in the real world.
Anthropic’s Claude AI has been praised and used by software engineers to check the code they write so they don’t make errors that cause a malfunction. Yet, you still need an expert who knows where to look and what to change.
Just look at where vibe coding is going for amateurs who initially claim they can create programs without any prior knowledge – it’s usually a fun hobby but the output isn’t good enough for a business.
Then there’s the question of cost. AI is not free because it takes up your time and it also costs money in terms of subscriptions (despite the massive “subsidy” by AI companies today to sign up users).
Personally, I’ve asked AI to help with the Techgoondu website – it has surprised me with the detail of instructions and the alternatives it offers but so far I have not been able to fix an issue with the width of images displayed (it makes images bigger but stretches the layout unevenly). Should I just pay an expert to do that?
What of the latest worry that AI will replace cybersecurity companies? The question to ask is whether you would trust an AI to manage your cyber defence, especially when you have little knowledge of the latest threats. Don’t forget: the bad guys are using AI to look for loopholes too.
And finally, what about AI taking away the jobs of Cobol software engineers? Well, if you are a bank, would you trust an AI to run your most fundamental systems with AI agents today? As a customer, you surely don’t want your balance wiped to zero because a rogue AI ran without guardrails.
Indeed, each week brings new warnings of AI’s limitations, even as it advances rapidly with each iteration. Witness how a top AI executive had her e-mail wiped out by an out-of-control AI agent. Or a company having its files erased by a hallucinating AI.
That is not to say that you should delay using AI. On the contrary, if you don’t try it out, you won’t know where the problems are integrating it into your business.
Coming back to Singapore’s AI push, this is why its focus has to be right. Already, there are many industry partnerships where people train with actual companies and get employed in the AI field. Or businesses get matched with AI startups to solve real problems.
Such programmes should expand and continue. The type of AI skills that will be relevant has to go beyond classroom theory or generic basics that apply generally. Those get outdated quickly.
Forget learning how to prompt. Learn how to make the AI work with the trusted data sources you use every day for work. Look for the loopholes and check where it falls short. Improve your work with AI, rather than let AI lead you by the nose.
The AI can do the brainwork by offering suggestions or even executing a command (if it’s good enough) but ultimately, a human in the loop has to be responsible.
At the end of the day, the one thing about AI that few of its most fervent proponents talk about is accountability.
To be accountable, humans have to know how AI works – they have to be able to explain how an AI has helped them come to a decision. In other words, the buck stops with them.
For Singapore, the type of AI skills learnt should be based on large-scale adoption across industries. Yes, there are templates to start with but AI is only as powerful as the data that powers it – that is the proprietary data on customers, sales and other operations that users need to harness to make AI work.
The reason why most of today’s AI deployments end up in failure is because they are driven by management and shareholders eager to see results without having a foundation in place.
Listening to the hype in the past two years, many have mandated AI projects that have ended up in failed experiments or worse, producing AI work slob when deployed without enough thought.
This is the gap that AI skills have to fill in Singapore. Turning the AI promise into reality, rather than hyping up its inevitability, is crucial. Having the AI literacy to know and check for AI errors should be basic.
AI is a means to an end, like learning to use a PC in the 1980s or more recently, learning to code. Already, people are asking why their kids are forced to code in school today, because AI can do that for you.
Well, yes, AI can do the scaffolding, maybe even 80 per cent of the way, but you have to debug your software, check it and stand by it as it powers your most important tasks. AI skills merely bring you to the starting line.
