Emerging markets are prioritizing top-line growth with agentic AI, despite the economic reality that hiring humans is still cheaper than running AI agents. This shift in focus is evident in the region's businesses, where projects driving top-line revenue are favored over those improving productivity and reducing costs. Arun Kumar Parameswaran, Salesforce's executive vice-president, highlights this trend during an interview with Computer Weekly. He notes that Singapore, a mature market, still has lower agent costs compared to human labor, but in other ASEAN markets, AI implementation is more cost-effective than hiring humans. Salesforce and other enterprises are deploying AI agents to unlock untapped revenue opportunities, such as sales development agents that handle lead scoring and appointment booking without human intervention. This trend is driven by the pricing economics in the region, where top-line use cases are prioritized over productivity improvements. However, many businesses struggle with the return on investment (ROI) of generative AI, with modest gains despite heavy investments. Srini Tallapragada, Salesforce's president, calls this the 'pilot purgatory' phase, where quick demos fail to translate into tangible financial impact. The crux of the ROI issue lies in the 'last-mile gap' - a lack of trusted data context, workflow integration, and guardrails to ensure AI agents' safety and effectiveness. Salesforce is addressing this by tracking agentic work units, a metric that measures actual tasks completed by agents, with over 2.4 billion units executed by customers in the fourth quarter alone. As AI projects move from pilots to production, the software industry must decide how to charge for autonomous agents, balancing consumption-based models with traditional licensing. Salesforce offers a mix of pay-as-you-go and agentic enterprise license agreements to accommodate customer needs. Lisa Singer, a market research analyst, emphasizes that the value of AI agents is not in usage volume but in the economic outcomes they enable, as seen in licensing models like AELA. The real bottleneck in agentic AI adoption is internal change management and business process reengineering, rather than infrastructure. As multi-agent ecosystems mature, enterprises will need a 'agent fabric' or control plane to govern interactions between AI agents from different suppliers, ensuring proper management and traceability.
