04/05/2026

BIZ & FINANCE MONDAY | MAY 4, 2026

19

Small models fuel future of enterprise AI agents

Bursa expected to trade range-bound this week

KUALA LUMPUR: Bursa Malaysia’s benchmark index is expected to trade in a range-bound mode with a cautious bias within the 1,700-1,730 range this week. Rakuten Trade Sdn Bhd vice president of equity research Thong Pak Leng said this is due to the absence of strong positive catalysts. He said volatility is likely to remain elevated as markets track developments in West Asia and oil price movements. “We maintain a cautious outlook on the domestic front, despite selective bargain hunting, as persistent geopolitical tensions and elevated crude oil prices raise concerns over inflation and cost pressures. “While energy-related stocks may find some support, broader market participation is expected to stay subdued as investors adopt a defensive stance,”Thong told Bernama. For the shortened week just ended, Bursa Malaysia traded mixed, with sentiment influenced by the West Asia conflict, US megacap earnings, the United Arab Emirates’ decision to exit the Organisation of the Petroleum Exporting Countries and its allies, and elevated oil prices. Bursa Malaysia and its subsidiaries were closed on May 1, in conjunction with the Labour Day public holiday. On a Thursday-to-Friday basis, the FBM KLCI added 1.68 points to 1,722.02 from 1,720.34 a week earlier. On the index board, the FBM Top 100 Index gained 10.76 points to 12,560.06, the FBM Emas Index increased 8.02 points to 12,723.74, the FBM Emas Shariah Index grew 131.75 points to 12,718.36, the FBM ACE Index fell 34.21 points to 4,618.11, while the FBM Mid 70 Index climbed 13.54 points to 18,085.97. By sector, the Financial Services Index sank 337.78 points to 19,885.03, the Industrial Products and Services Index added 3.95 points to 196.31, the Energy Index increased 10.36 points to 840.67, and the Plantation Index improved 94.05 points to 8,939.31. Weekly turnover slipped to 14.15 billion units valued at RM13.03 billion from 16.39 billion units valued at RM15.16 billion a week earlier. The Main Market volume decreased to 8.86 billion units valued at RM111.94 billion from 9.33 billion units valued at RM13.66 billion previously.

IN the AI industry, we’ve spent the last 3 years obsessed with scale. We’ve chased parameter counts into the trillions, believing that “bigger” was the only path to “smarter.” But as the dust settles, a new reality is emerging for the enterprise-size is not the metric that matters, delivering reliable, deterministic outcomes is. At Red Hat, we’ve always believed that the most powerful technologies are those that are distributed, open, and fit-for-purpose. Small language models (SLMs) represent that exact shift. The distinction between SLMs and large language models (LLMs) is less important than the architectural role the model serves. What matters is the functional sovereignty a small model brings to the table. We are moving away from a world of conversational AI—where we ask a giant, black box model a question—and entering the era of agentic AI, where a fleet of specialised models performs the actual work of the business. Every business will run AI agents We are on the verge of a shift as fundamental as the transition to the web. Think back to the evolution of business identity. In 1995, the industry asked, “Why do I need an email address?” In 2005, it was a website. In 2015, a social media presence. In 2026, the question will be, “How many agents do I have running?” We are heading towards a world where there will be more AI agents than people. Every business will have a swarm of them: 0 Customer-facing agents that don’t just answer questions, but solve complex logistics issues. 0 Workflow agents that automate the invisible “glue” between departments. 0 Headless agents that silently execute API calls to reconcile inventory and process payments. But you cannot build a sustainable, cost effective agentic fleet on someone else’s subsidised cloud tokens. This is where the SLM becomes the mandatory tool to enable enterprise use cases and scale. Why SLMs rule the agentic backend While frontier LLMs are masterpieces of high throughput engineering, they are often too heavy for the role of a reflexive digital employee. In an agentic workflow, we don’t just need raw power, we need low-latency execution. SLMs allow us to provide sub-second response times and deterministic reliability that business critical automation demands.

Weeks

Noriega

1. The power of specialisation (efficiency > scale) While few organisations would consider fine-tuning a 400B-parameter model, a 3B or 7B model offers a manageable and highly effective entry point. This is where architectural control begins. Research from late 2025 demonstrates that even a 350M-parameter model fine-tuned on high-quality, synthetic data can outperform generalist frontier models in specific tool calling and API-orchestration domains. For a robust agentic backend, the goal isn’t broad, poetic language capability—it is high-precision specialization. 2. Determinism and the “math of reliability” One of the biggest hurdles for enterprise AI is non-determinism, the risk that an agent might format a response correctly one time and fail the next. While no LLM is a perfectly deterministic math function, SLMs allow us to enforce architectural control that was previously much harder. By using constrained decoding techniques like JSON Schema or Context-Free Grammars (CFGs), we can prune the model’s token search space, making it physically impossible for the model to choose an invalid next character. This shifts the focus from open-ended magic to schema-constrained accuracy. Combined with local execution and specialised fine-tuning, SLMs can achieve over 98% validity in structured tasks, offering the predictable reliability required for sensitive agentic workflows. 3. Data sovereignty is not optional Your data is your most precious asset. In an agentic world, these models will handle your customer relationship management (CRM),

your proprietary code, and your internal strategy. Giving that data away to a third-party cloud provider in exchange for “intelligence-as a-service” is a strategic mistake. Running SLMs on-prem or within your own hybrid cloud environment means you remain the owner of your IP. It allows for a “zero trust” AI architecture where sensitive data never leaves your perimeter, fulfilling the strict regulatory requirements common in industries such as healthcare, finance, and government. Final thoughts We are transitioning from a world of generative AI (gen AI) producing conversation and content to one of agentic AI taking action on our behalf. In this new era, the question is no longer about which model is the biggest, but which infrastructure is the most reliable and protected. When your business operations depend on a fleet of specialised digital agents, the “black box” cloud model is no longer enough. You need sovereignty, speed, and precision. At Red Hat, we believe the path to the agentic future is open. By leveraging curated small language models that can be fine tuned, served, and orchestrated with the Red Hat AI portfolio, enterprises can move AI out of the lab and into the core of their business logic. The space is moving fast, but the goal is clear: stop chasing the giants and start building the backbone. The future of AI is small, fast, and built on the open hybrid cloud. This article is contributed by Red Hat AI engineering director Catherine Weeks, and initiative tech lead @ emerging technologies – office of the CTO Ricardo Noriega. for growth. What we are witnessing today is not just technological progress, it is the emergence of a new digital culture where intelligence is embedded into the devices we use every day. As AI becomes more predictive and autonomous, the responsibility to develop it thoughtfully becomes even more important. At Samsung, we believe AI must always remain human-centric, secure, and ethical. Technology should empower people, while respecting their privacy, choices, and control. Malaysia’s openness to innovation has always been one of its greatest strengths. As we move further into the AI era, Samsung will emulate this spirit to shape how technology is adopted, adapted, and ultimately used to improve lives. The future of AI will not simply be defined by what technology can do, but by how it helps people realise their full potential. This article is contributed by Samsung Malaysia Electronics president Charles Kim ( pix ).

Malaysia’s next digital chapter: Embracing the age of agentic AI MALAYSIA is standing at a defining moment in its digital journey. Across industries, classrooms, and homes, artificial intelligence (AI) is no longer seen as a distant innovation. Instead, it is becoming an everyday tool that helps Malaysians work smarter, communicate more effectively, and express their creativity in new ways. responding to prompts and begins to actively assist users in meaningful ways. With the Galaxy S26 Series, Galaxy AI is designed to reduce friction in everyday experiences, allowing people to focus on what matters most.

Innovations such as Now Nudge, alongside experiences like Now Brief and Now Bar, move Galaxy AI closer to becoming a true AI companion. Rather than functioning as a collection of tools, Galaxy AI is evolving into an intelligent system that understands context, learns routines, and surfaces relevant information at the right moment. The goal is simple: to make technology feel more intuitive, helpful, and seamlessly integrated into daily life. For Malaysians, this evolution is especially meaningful. As a digital nation that thrives on connectivity, collaboration, and entrepreneurship, AI has the potential to empower individuals and communities alike. From students exploring new ways to learn, to creators expressing ideas more freely, to professionals navigating increasingly digital workplaces, AI can unlock new opportunities

This shift reflects something deeper than the adoption of new technology. It signals a shift to embrace intelligent systems that anticipate needs, understand context, and support people in navigating an increasingly fast-moving world. The response to the recently launched Samsung Galaxy S26 Series reflects this momentum. With strong global demand and more than 1.3 million pre-orders for the Galaxy S26 Series, it is clear that consumers are increasingly looking for devices that go beyond performance, they want technology that works intelligently alongside them. At Samsung, we believe we are entering the agentic AI era, where AI evolves beyond simply