As enterprises across Asia accelerate investments in AI agents and workflows, the conversation is rapidly shifting beyond model performance. The challenge is moving to scaling AI without allowing hidden operational costs, governance complexity and technical debt to stymie business value.
Ed Keisling, Chief AI Officer at Progress Software, believes the next phase of enterprise AI adoption will be defined less by model capability than by an organisation's ability to build trusted retrieval systems, reusable knowledge layers and governance that can support AI at scale.
Speaking with iTNews Asia, Keisling said the biggest barrier to sustainable AI returns will not be model performance, but the operational costs that emerge as deployments expand - from token consumption and infrastructure overhead to retrieval inefficiencies and human oversight.
Hidden costs often emerge long after successful pilots
According to Keisling, one of the misconceptions is assuming an AI system that performs well in demonstrations will naturally scale across an enterprise. Many organisations continue to rely on custom governance and observability layers to manage AI agents. While workable for limited deployments, he said these bespoke approaches struggle as enterprises begin orchestrating thousands of agents operating across multiple AI models and vendors.
While emerging enterprise control platforms are beginning to centralise governance, security and observability, organisations still need time to evaluate and migrate existing deployments.
As many organisations celebrate successful AI proofs of concept, Keisling cautions that the largest costs often appear only after production deployment. He explained that early pilots typically involve limited users, datasets and governance requirements, masking the operational demands that emerge as AI agents become more autonomous. At enterprise scale, retrieval inefficiencies, reasoning loops, infrastructure consumption and ongoing monitoring quickly compound into significant operational expenditure.
"Teams tend to assume that once an agent works in a demo, it will scale cleanly. In practice, poor data readiness and loosely governed retrieval amplify errors, forcing organisations to invest later in rework, tuning and remediation that could have been avoided," Keisling said.
He also explained that enterprises often underestimate how quickly costs become unpredictable in agentic architectures. “As agents take on more autonomy, inefficiencies in retrieval, reasoning loops or data preparation compound into real spend through token waste, latency and infrastructure overhead." he added.
Poor retrieval is becoming an expensive business problem
Rather than viewing AI challenges purely as a model problem, Keisling argued that enterprises should pay greater attention to retrieval architecture.
Traditional retrieval approaches often rely on loading large volumes of enterprise documents into model context windows and expecting AI models to determine relevance independently. While this may produce acceptable demonstrations, he said it creates inconsistent outputs, higher operating costs and governance risks when deployed at scale.
Agentic Retrieval-Augmented Generation (RAG) changes that equation by making retrieval goal-driven, structured and continuously validated. Instead of simply retrieving information once, agentic systems iteratively refine retrieval, validate relevance and ground responses using approved enterprise knowledge.
The goal of an agentic system is to enable it to plan and act, not just to generate answers.
- Ed Keisling, Chief AI Officer, Progress Software.
He added that this structured retrieval process also shifts data quality efforts upstream by forcing organisations to define business problems first before preparing the data needed to solve them.
Shared knowledge layers can help reduce long-term costs
For CIOs and CFOs, Keisling said investment decisions should increasingly focus on reusable enterprise knowledge rather than individual AI applications.
He urged organisations to create standardised retrieval pipelines capable of supporting multiple assistants, automation initiatives and search use cases from a common governance foundation. Without that shared layer, every new AI deployment risks duplicating engineering effort while increasing governance complexity.
"CIOs and CFOs need to realise the value of an investment that builds towards a standardised enterprise knowledge layer that will enable ROI through the reuse of retrieval pipelines across use cases,” he said.
He explained that such an approach also reduces the risk of outdated, unapproved or sensitive information being surfaced by AI systems, improving both consistency and enterprise trust.
Operational risk can extend beyond infrastructure
Keisling said rising operational costs are only part of the enterprise AI challenge. As organisations introduce greater levels of AI autonomy, balancing agent independence with governance becomes increasingly difficult.
Errors can multiply quickly, while limited visibility into agent decision-making makes troubleshooting and auditing far more complex. He also warned against excessive automation of decisions that continue to require human judgement and business context.
Governance must be measurable
According to Keisling, agentic RAG strengthens compliance and auditability by grounding outputs in permission-controlled enterprise knowledge, maintaining retrieval logs and providing traceable citations.
Rather than relying on trust alone, organisations gain evidence showing what information was retrieved, how decisions were made and whether responses remain reliable over time.
He said evaluation metrics, observability and continuous monitoring will become essential capabilities as AI systems expand across enterprise operations.
For many mid-market businesses across Asia-Pacific, Keisling believes SaaS based agentic RAG offers an opportunity to avoid building increasingly complex AI infrastructure internally.
Instead of maintaining large engineering teams for governance, retrieval and ongoing optimisation, organisations can accelerate deployment while supporting multiple AI use cases through a common enterprise knowledge foundation.
He said organisations also benefit from improved token efficiency, multilingual capabilities and built-in governance that helps simplify increasingly diverse regulatory requirements across the region.
Enterprise AI success will be measured by business outcomes
Looking ahead, Keisling expects agentic RAG to become a standard capability rather than a competitive differentiator over the next few years. "Successful organisations make AI 'boring' in the best possible way - predictable, testable, observable and governed,” he said.
To be successful, he said the organisations must generate measurable returns and redesign their workflows rather than simply layering AI onto existing processes. They must ground AI in trusted enterprise data, establish human oversight, implement cost guardrails and continuously evaluate business outcomes.
”The enterprises that treat governance, retrieval and operational discipline as strategic capabilities, not technical afterthoughts will be best positioned to turn AI into sustained business value rather than mounting operational expense,” Keisling said.
