Will we be the last generation of a purely human workforce? That was the implicit question posed by GovTech’s Chang Sau Sheong at his opening keynote at Apidays Singapore 2025. Chang was paraphrasing remarks made by Salesforce CEO Marc Benioff at the World Economic Forum in January, in a call to reskill workers for the road ahead.
And this isn’t hyperbole. AI is changing how we work, even for the technical experts and software developers at the forefront of technology. They turned out in droves to fill every conference session at Marina Bay Sands (MBS), eager to understand AI’s impact on APIs and how to build tomorrow’s intelligent ecosystems.
From narrow tools to thinking agents
Where are we with AI today? Chang traced the rapid development of AI for the standing-room-only crowd: “Only recently, three or four years back, we were still using very specialised AI, like OCR, facial recognition, and computer vision. Then things changed rapidly when foundation models came about.”
“Suddenly, AI became more general-purpose, faster, and more adaptable. That led quickly to generative AI, where AI started creating things, no longer just recognising or predicting or classifying,” he said, adding that agentic AI is what’s next.
Chang noted that agentic AI goes beyond traditional automation, with agents that can operate without direct human guidance. Agentic AI can handle more complex tasks and perform a much broader range of functions than the narrow and specialised AI we've seen so far. “It’s still a bit stumbling, a bit shaky, but we’re clearly moving into a world of greater autonomy and reasoning. AI is starting to become aware of its environment.”
But should programmers fully rely on AI tools to code? This is where the picture becomes murkier. White coding – or using AI to code – is appealing due to its promise of greater productivity, potentially turning even an average engineer into a sough-after “10x engineer."
But there’s pushback, with some urging engineers to learn to code the traditional way instead. Chang offered an observation on this front: “A lot of research papers on AI coding are written by big companies. They’re not just using off-the-shelf co-pilots; they’re using internal tools. The secret sauce.”
The hidden risks of AI
In his keynote, Manjunath Bhat, VP Analyst at Gartner, explored how AI is reshaping software engineering. He spoke about multi-agent systems, the advent of the AI-native software development life cycle, and the rise of AI gateways. Like the aeroplane, AI is so disruptive that it will reinvent many of today’s business models, he noted.
One emerging concern is the security risks posed by AI systems. “If you ask any CISO their biggest concern today, it is essentially: shadow AI. You’re bringing in these agents that the security team has no visibility over. Every MCP (Model Context Protocol) server that you bring in is a blind spot for the C-suite.”
Where human users are traditionally assigned privileges with clear segregation of duties, this is rarely the case with AI agents. Moreover, humans assigned to manage or work with agents gain implicit access to their privileges. This means that people who interface with multiple agents effectively gain elevated privileges, he pointed out.
Finally, should developers white code? Bhat left the question open but used a comic strip to deliver a pointed warning: two engineers can now produce the output of 50 using vibe coding – while also generating the technical debt of 50. “If you're an engineer, send this message up the hierarchy: ‘You can ask me to white-code, but someone will have to pay off this debt someday.’”