AI Agents Powered by SLMs Will Drive Digital Economy of Future LLMs excel at broad knowledge tasks, creative content generation, and data reasoning but they come with significant costs, says Vishal Chahal Vice President, IBM India Software Labs

Today's digital economy is evolving rapidly, shaped by a growing web of microtasks and interactions fueled by AI and automation. As AI becomes mainstream, the challenges of scaling, programming AI models for enterprise usability, and maintaining human oversight will only intensify. Enterprises are now seeking targeted AI solutions. This demand is ushering in a new era powered by AI agents.

The Growth Of Agentic Revolution

AI agents represent a major shift, as they don't just interact. They understand how tasks link together, what data must be exchanged, and how to adjust actions as circumstances change. When faced with repeated tasks, these agents recognise variations, adapt dynamically, and even create new connections on the fly using GenAI models, such as IBM's Granite. The outcome? Enterprises need less IT intervention as AI agents orchestrate tasks independently, scaling on demand. Gartner predicts that by 2028, 33% of enterprise software applications will embed Agentic AI, a major leap from less than 1% in 2024, enabling 15% of day-to-day work decisions to be handled autonomously.

Evolving Small Language Models

Small Language Models (SLMs) are compact yet powerful models ranging from 100 million to 7 bil lion parameters. Unlike Large Language Models (LLMs), SLMs are purpose-built, consuming far fewer resources while delivering remarkable accuracy, and helping reduce infrastructure costs by 60-80%. Financial services firms that have switched to SLM-based agents for fraud detection are already witnessing results, with faster processing times and more accurate results. SLM-powered agents can continuously monitor inventory, predict disruptions, and automatically adjust procurement schedules, reducing stockouts by 20-30% and decreasing inventory carrying costs by 15-20%. All this without the need for an expensive, large-scale AI infrastructure.

Addressing Safety and Governance

The growing independence of AI Agents raises critical questions about safety and governance. How do we ensure that agents act responsibly without slipping into unintended actions? Responsibility demands that backend models are fortified, tuned to prevent hallucination or eliminate bias. Reflection ensures that agents not only act but assess as well, constantly evaluating whether better, safer actions exist under changing conditions. Building robust governance frameworks are non-negotiable as agents are becoming integral to enterprise operations.

LLMs Versus SLMs

LLMs excel at broad knowledge tasks, creative content generation, and data reasoning. But they come with significant costs, given that they are both infrastructure-heavy and resource intensive. SLMs, by contract, offer a leaner and agile alternative. Trained on focused datasets, they perform specific processes with remarkable speed and minimum computational demand, capable of responding in milliseconds. SLMs can also operate on edge devices and standard enterprise hardware, eliminating the need for specialised GPU infrastructure.

SLMs and AI Agents In Digital Economy

As AI becomes ubiquitous, specialized SLMs paired with intelligent Agents that are purpose built will form the operational backbone of the digital economy. The future will belong to the smartest, most appropriately-sized, and well-governed system, working tirelessly within an agile agentic framework with the right balance of autonomy and oversight.