Author: Agentic AI Systems C334

At a hawker stall, paying for lunch can be as simple as passing over a few notes—or scanning a QR code on the counter. By November 2024, digital payments had become familiar at many Singapore hawker stalls, but cash had not disappeared. The Straits Times reported that more than 11,500 hawkers had begun accepting payments through SGQR, while hawkers in eight food centres still accepted cash only.

For some stallholders, sticking with cash was a practical choice. Angie Sim, a 65-year-old noodle-stall owner at West Coast Food Centre, said cash was easier for her and suited many of her mostly older customers, who were less comfortable with technology and carried cash. Other hawkers described difficulties learning or navigating digital platforms. Poor eyesight, concerns about inconveniencing customers and the need to check whether payments had gone through were also reasons for hesitation.

One drinks-stall owner faced a different challenge: customers sometimes arrived without cash. He sometimes served them and let them return to pay later. These accounts show that payment methods are not just a question of technology. They also affect how hawkers serve customers and manage a busy stall.

For hawkers who had adopted SGQR, digital payments could make work easier. Madeline Chen, who runs a braised-pork stall at Chinatown Complex Market & Food Centre, said they reduced the work of handling cash when she was busy. Some regular customers even paid before collecting their food. At Bendemeer Market & Food Centre, prawn-noodle seller Lim Yaw Fook said his sales had risen by about 20 per cent since he began using SGQR in 2020. He associated the increase with more younger customers visiting his stall.

The government’s Hawkers Go Digital programme, launched in 2020, was designed to support the shift. Under the arrangement described in the article, the 0.5 per cent transaction fee was waived on the first $20,000 of monthly SGQR transactions. Enterprise Singapore and the Infocomm Media Development Authority announced on October 10, 2024, that the fee waiver would be extended to the end of 2025. The article described this as part of a push to encourage the 18,000 stallholders in government-run food centres to adopt digital payments.

The picture in the article is not of one payment method replacing another overnight. Instead, QR codes and cash sit side by side, with hawkers weighing convenience, customer needs and the practical demands of running a stall. For diners, carrying cash may still be useful; for stallholders, the right way to take payment can depend on the people and circumstances at the counter.

Source: Why cash is still king for some hawkers in Singapore despite digitalisation push, The Straits Times

Imagine an AI system that does more than answer a question. It reads an email, decides what needs to happen, updates a spreadsheet and asks someone for approval. That kind of multi-step task is what the Intellisoft Training guide describes as agentic AI—and learning to build or use it can take different paths.

The guide’s main message is that there is no single course that suits everyone. A good choice depends on your technical experience, what you need AI to do at work and which tools you already use.

For people without programming experience who want to automate repetitive tasks, the article points to a two-day, 16-hour no-code course, using “Create Agentic AI Automations Without Coding” as an example. It lists an approximate fee of S$390 after 70% funding. The guide also describes a two-day course using Claude Code and OpenAI Codex for analysts, power users and developers who want to build AI applications and agents; its approximate post-funding fee is S$468.

The right tools may also depend on your workplace. Staff whose work centres on Microsoft 365—such as Outlook, Excel, SharePoint and Teams—might find Power Automate a better fit. Those who also want to build business applications could consider Power Apps alongside Power Automate. For IT staff and power users creating custom copilots, the guide names Microsoft Copilot Studio and Microsoft 365 Copilot Agents, while noting that these options are not yet funded in the guide.

Not everyone needs to build an agent. If your main goal is to write, research or analyse information more effectively with ChatGPT, Gemini or Copilot, the article suggests that a general AI-tools course—or free self-paced tutorials—may be enough. And for learners who want to understand the code behind AI applications, it suggests learning Python fundamentals before taking an AI coding course.

Cost is another part of the decision, but the figures need careful attention. The guide says eligible SkillsFuture or WSQ-funded courses have a TGS course reference. It gives examples of 70% funding for Singapore Citizens aged 40 and above and SME-sponsored staff, and 50% for other eligible Citizens and Permanent Residents. SkillsFuture Credit, SFEC and UTAP may further reduce the remaining cost. These amounts depend on eligibility; the listed fees are not guaranteed for every learner.

There is also a useful detail to keep in mind when weighing the recommendations: Intellisoft says it runs several of the courses featured in its own guide. Its suggestions are the publisher’s guidance, not an independent ranking. The article does not establish which course is objectively “best,” compare teaching quality or learner outcomes in a controlled way, or show that a course leads to a particular job or salary.

So the most practical first step is to define your goal: automate routine tasks, build applications, work within Microsoft tools or simply get more from everyday AI assistants. Then check the course content, tools and funding eligibility against that goal.

Source: Best Agentic AI Course in Singapore (2026): How to Choose, Intellisoft Training

The dot-com era offers a useful lens on today’s AI boom—but not a crystal ball. In an August 2025 Forbes article, Harvard research fellow Paulo Carvão argues that history points to genuine risks without proving that an AI crash is imminent.

One clear parallel is market concentration. At the time of publication, the “Magnificent Seven” represented more than a third of the S&P 500. At the dot-com peak in 2000, five prominent technology stocks made up 15% of the index. Such concentration can make markets vulnerable, Carvão argues, but it does not by itself mean a crash is inevitable.

The infrastructure buildout offers another caution. During the dot-com boom, companies invested heavily in fiber-optic networks before demand arrived quickly enough to support the spending; bankruptcies followed. Today, leading AI companies are pouring money into data centers, raising a familiar question: will demand justify the investment?

Yet the earlier buildout also left behind infrastructure that later helped support the internet economy, application businesses and cloud computing. Carvão suggests AI data centers could likewise prove useful over time—but presents that as a possibility, not a guaranteed payoff.

The strongest case for AI’s broader significance may come from its adoption. The article cites a National Bureau of Economic Research survey reporting that, by late 2024, about 40% of the U.S. population had used generative AI, while 23% had used it for work at least once in the previous week. The study also found that workplace adoption was progressing faster after launch than adoption of personal computers or the internet.

Those figures point to AI’s potential to become a consequential, general-purpose technology. They do not settle whether today’s valuations are justified or whether data-center investments will earn adequate returns.

That is the careful lesson in Carvão’s comparison: the dot-com story is a warning about concentrated markets and infrastructure built ahead of demand, not a timetable for an AI collapse. Even if a market correction comes, the technology’s lasting value could outlive it.

For years, polytechnic education in Singapore was seen as the fallback route — the path taken by students who couldn’t make it into junior college (JC). But that narrative is shifting.

According to a recent feature by CNA’s TODAY, the country’s five polytechnics — Ngee Ann Polytechnic (NP), Singapore Polytechnic (SP), Republic Polytechnic (RP), Nanyang Polytechnic (NYP), and Temasek Polytechnic (TP) — have undergone a remarkable transformation in both reputation and relevance. With around 20,000 students enrolling across these institutions each year, polytechnics are no longer viewed as a second-choice option but as a compelling and capable pathway to careers and higher education.

### Applied Learning Meets Real-World Experience

One of the primary drivers of this change is the emphasis on applied learning. Polytechnics today offer a wide range of courses — some even unique in Singapore. Republic Polytechnic, for example, offers a sonic arts programme that is exclusive to the institution. These specialised programmes attract students with specific passions and ambitions that a more traditional academic route might not serve.

Beyond the classroom, polytechnic students engage in industry talks, real-world projects, and internships that bridge the gap between theory and practice. The article highlights a Temasek Polytechnic graduate who later pursued further studies at the National University of Singapore (NUS), demonstrating that the polytechnic pathway can lead to top-tier universities just as well as the JC route.

### Stronger Industry Links as a Key Advantage

Another major factor in the rising appeal of polytechnics is their close collaboration with industry. These partnerships ensure that students are exposed to current practices and emerging trends, giving them a head start in the job market upon graduation. Outreach efforts by the polytechnics have also played a role in reshaping public perception, showing students and parents alike that a polytechnic education can be both rigorous and rewarding.

### Challenges on the Horizon

However, the article doesn’t shy away from the challenges that remain. One significant issue is the wage gap between university and polytechnic graduates. Despite the growing respect for polytechnic education, university degrees still command higher starting salaries on average. There is also the pressing risk that skills acquired in polytechnic courses could become obsolete as industries evolve at breakneck speed.

To address these concerns, the polytechnics are doubling down on collaboration with industry partners, ensuring that curricula stay relevant and that students are equipped with adaptable, future-ready skills.

### A Long Way, But Room to Grow

The feature by CNA’s TODAY suggests that Singapore’s polytechnics have indeed come a long way — shedding the stigma of being an “inferior option” and carving out a distinct identity grounded in hands-on learning and close ties to the working world. Yet, there is still room to grow. Whether it’s closing the wage gap, anticipating the skills of tomorrow, or forging even deeper links with industries, the next chapter for polytechnics will require continued innovation.

For students deciding on their post-secondary pathway, the message is clear: polytechnic education isn’t a consolation prize — it’s a powerful launchpad in its own right.

When most people think of competition in the tech industry, they picture head-to-head product battles: Coca-Cola versus Pepsi, Uber versus Lyft, or Intel versus AMD. But as a new article from *California Management Review* argues, the most consequential rivalry shaping the future of technology is not a direct product war at all. It is a clash of sprawling, interconnected systems — and the two titans at the center are NVIDIA and Apple.

Published on October 1, 2026, the article titled “Capabilities, Supermodularity, and Competition: Insights into the Future Fortunes of NVIDIA vs Apple” reframes the relationship between these two California giants. They are not simply competing to sell better chips. Rather, they are competing over which company’s collection of connected assets — hardware, software, developer ecosystems, and customer relationships — will capture the lion’s share of value as artificial intelligence spreads through every corner of the economy. ([cmr.berkeley.edu](https://cmr.berkeley.edu/2026/10/capabilities-supermodularity-and-competition-insights-into-the-future-fortunes-of-nvidia-vs-apple/))

## Two Very Different Empires

At first glance, NVIDIA and Apple barely seem like rivals. NVIDIA’s core business is supplying AI infrastructure to data centers, cloud providers, and enterprises. Its chips power the large language models and generative AI systems that have become central to modern computing. Apple, by contrast, sells premium consumer devices — iPhones, iPads, Macs — along with software, services, and an increasingly important ecosystem of wearables and subscriptions.

Yet the authors argue that focusing on these surface-level differences misses the deeper strategic picture. The real contest is about which system — NVIDIA’s AI-compute stack or Apple’s consumer-device ecosystem — will become the dominant platform through which AI delivers value to users. ([cmr.berkeley.edu](https://cmr.berkeley.edu/2026/10/capabilities-supermodularity-and-competition-insights-into-the-future-fortunes-of-nvidia-vs-apple/))

## The Power of Complementary Assets

The article applies the “Profiting From Innovation” framework to explain why this matters. A chip’s value, the authors point out, is never just about the silicon. It depends heavily on the capabilities surrounding it: software libraries, developer relationships, manufacturing access, distribution channels, and customer connections. These complementary assets can make or break a technology’s commercial success.

NVIDIA’s real advantage, the article argues, is not simply that it makes powerful GPUs. It is that it has built an entire system around those chips — most notably CUDA, its parallel computing platform and software ecosystem, along with deep developer adoption, optimized libraries, and carefully cultivated supply-chain relationships. Competitors like AMD, which makes competitive GPUs, and Intel, which is investing heavily in AI accelerators, have struggled to replicate this ecosystem. As the article notes, competing hardware alone is not enough. ([cmr.berkeley.edu](https://cmr.berkeley.edu/2026/10/capabilities-supermodularity-and-competition-insights-into-the-future-fortunes-of-nvidia-vs-apple/))

## Apple’s Countermove

Apple’s competitive position rests on a very different but equally powerful set of assets. The company commands a massive, loyal user base, an integrated hardware-software ecosystem, and deep relationships with app developers and content creators. These are formidable complementary assets in their own right.

But the article also highlights a fascinating area of emerging overlap: Apple is developing its own AI server chip for inference in its data centers. This chip is intended to reduce Apple’s reliance on NVIDIA hardware for cloud processing tied to Apple Intelligence and future Siri capabilities. It is a quiet but significant move — a signal that Apple sees the AI infrastructure layer as strategically important, even if its primary business remains consumer-facing. ([cmr.berkeley.edu](https://cmr.berkeley.edu/2026/10/capabilities-supermodularity-and-competition-insights-into-the-future-fortunes-of-nvidia-vs-apple/))

## Systems, Not Silos

Ultimately, the article’s central insight is that the most important technology competition of the coming decade will not be fought on a single battlefield. It will be a contest between two comprehensive systems, each with its own strengths, its own customer base, and its own vision for how AI should integrate into daily life.

NVIDIA’s system is optimized for the data center — for training massive models and running inference at scale for enterprises and cloud providers. Apple’s system is optimized for the individual — for delivering AI-powered experiences directly to consumers through devices they already own and trust.

As AI continues to diffuse through the economy, the question is not which company makes the better chip. It is which connected system of capabilities will capture more of the value that AI creates. And that, the article suggests, is a much more interesting — and uncertain — contest than any simple product comparison could capture. ([cmr.berkeley.edu](https://cmr.berkeley.edu/2026/10/capabilities-supermodularity-and-competition-insights-into-the-future-fortunes-of-nvidia-vs-apple/))

In the article “The Evolution of Agentic Harnesses: From Chains to Graph to Code” (InfoQ, October 26 2026), a clear architectural journey unfolds—one that mirrors the maturing ambitions of agentic AI systems. At its core are three structural paradigms: chains, graphs, and code harnesses.

First, chains are the simplest of the three. They string prompts together in a single, linear sequence—quick to implement and easily tuned for narrowly scoped tasks. The article underscores their vulnerability: a misstep early in the chain can cascade, undermining the entire workflow.

Graphs deliver a more expressive structure. With nodes, edges, branching paths, and parallel execution, they excel in complex, multistep scenarios. Not only can these harnesses adapt dynamically to different paths, they also support better state tracking and recovery when something goes wrong.

At the most sophisticated level, code harnesses treat AI execution as a first-class, programmable construct. By incorporating sandboxed environments and iterative refinement loops, they enable tighter control, verifiability, and correction of model outputs—mitigating the risk of relying on untrusted or unexpected AI behavior.

The article doesn’t posit that every agent should default to the most advanced paradigm. Instead, it offers engineering wisdom:

• Treat all model outputs as inherently untrusted, and place rigorous validation or verification around their use.
• Separate planning from execution—allowing clear, auditable reasoning before action.
• Build visibility into every step (observability) to diagnose issues and ensure reliability.
• Start with the least complex architecture that satisfies your requirements (be that speed, cost, or trustworthiness).

Viewed as a spectrum, chains represent ad hoc convenience, graphs offer structured robustness, and code harnesses embody production-grade resilience. The progression reflects how real-world demands—reliability, error handling, observability, and correctness—drive the adoption of more formal and controlled agent architectures.

This evolution doesn’t imply a rigid linear path: there are no benchmarks here showing chains always giving way to graphs, or graphs superseding code harnesses across the board. Rather, InfoQ presents a design philosophy—choose the right tool, at the right time, balancing simplicity with the rigors of production-quality engineering.

In the grand arc from chains to graphs to code, we see the maturing of AI agent design: a shift from quick-and-dirty prompt workflows toward modular, verifiable, maintainable, and robust systems. For practitioners charting a course through this landscape, the article offers both a map and a compass—not a mandate, but a mindset for building trustworthy intelligent systems.