SINGAPORE — Rising token costs, weak data foundations and uncertainty over which tasks are suitable for artificial intelligence are becoming major obstacles as enterprises move AI from pilot projects into production, according to Boomi Chief Technology Officer for Asia Pacific and Japan David Irecki.
Boomi, is an enterprise technology company focused on data activation and AI. Its Boomi Enterprise Platform brings together integration and automation, data management, API and MCP management, and AI capabilities to help organizations connect systems and manage AI agents.
In an exclusive interview with The Manila Times on Sept. 10 during the Boomi World Forum regional media event in Singapore, Irecki said businesses are increasingly questioning whether the cost of running AI agents is justified by their results.
“I think the use case forms a part of that: understanding was the use case the right use case for AI,” Irecki said during the interview. “But equally, for those that are moving into production, it’s that token cost is becoming probably the biggest discussion at the moment, because businesses are looking at it and going, ‘Well, is the investment in token cost worth the outcome?’”
He said some Boomi customers in the Asia-Pacific region have rolled back AI agents after determining that deterministic processes or human intervention were more appropriate for particular tasks.
“So I think it’s a multi-layered conversation at the moment,” Irecki said.
His comments followed Boomi’s Sept. 2 announcement of new capabilities centered on its Agent Control Plane, which the company describes as an AI-native infrastructure layer intended to govern AI agents, control costs and connect agents securely to enterprise systems.
The company said the Agent Control Plane is designed to provide visibility over agents and tools, enforce policies, control token spending and require human approval for high-risk transactions. It can operate across public cloud, private cloud and on-premises environments.
Cost, model choice
Irecki said concern over token costs is also driving interest in bring-your-own-model approaches, open-weight models and small language models.
“As organizations are looking for choice and flexibility, if they can have a free capability on-premise that they can wrap in the context of their business, that’s very advantageous,” Irecki said
He said an AI gateway could route simpler requests to local models while sending more complex workloads to frontier models such as Claude or ChatGPT.
In his keynote at the forum, Irecki said enterprises are increasingly scrutinizing the return on AI investments and the proprietary business context they provide to frontier-model providers.
“There’s one thing greater than AI, and that’s ROI,” he said on stage.
Data remains
a foundation
Irecki said enterprises also need to assess data quality and integration readiness before deploying agents broadly.
Some AI projects work during the pilot phase because they use manually curated datasets and narrowly defined connections to back-end systems, he said. Problems can emerge when those systems are expanded across the wider organization.
“We had one customer that scaled that out and very quickly saw hallucinations and inconsistencies because, to your point, that narrow path they had integrated properly and got the data quality right, but then it exposed that the rest of their organization wasn’t,” Irecki said during the interview.
He recommended starting with the desired business outcome and working backward to determine which systems need to be connected, what data is required and how it can be securely exposed to AI.
In his forum keynote, Irecki similarly emphasized that integration, data management and API management remain foundational as enterprises adopt AI agents.
Philippines,
Southeast Asia
For the Philippines, Irecki said legacy modernization remains part of the technology discussion, but companies do not necessarily need to replace their existing IT infrastructure before adopting AI.
“Of course! Yeah, we still connect mainframes today,” Irecki said during the interview, noting that organizations can connect modern APIs as well as databases and flat files.
He said some companies that moved quickly into AI are now reassessing their investments after failing to achieve expected outcomes, while companies that spent more time modernizing their technology foundations may be able to adopt AI more selectively.
Irecki also sees AI as an augmentation tool for industries such as business process outsourcing.
“We’re really seeing AI more and more become a copilot to the human,” he said during the interview. He said some human skills, knowledge and intuition remain difficult to reproduce in AI agents, making human oversight important even as automation expands.
“For something like the BPO industry, I think it’s an accelerator to what they do and the services they can offer,” he said.
He added that regional differences in regulation, data-sharing practices and language could affect AI deployment. Locally adapted models could eventually incorporate Philippine or other Southeast Asian languages, business practices and cultural norms, he said.
Governance,
gradual adoption
Irecki said governance needs to be considered alongside technology deployment rather than after an AI system is already in production.
Boomi’s announcement said its Agent Control Plane is intended to provide a centralized governance layer across different AI agents, models and applications. Its AI Gateway is designed to provide observability, identity and policy controls, model routing and rate limiting.
On stage, Irecki said enterprises need visibility into their agents, control over access to systems and the ability to determine which models handle particular workloads.
For smaller companies, he recommended a gradual approach.
“You can look at what’s key to connect; you don’t have to do a big bang project,” Irecki said during the interview. He added that open-weight models and small language models could provide cheaper entry points as their capabilities develop.
At the close of his presentation on stage, Irecki said enterprises have moved beyond initial AI experimentation and now need to determine where the technology makes business sense.
“The days of pilots are gone,” he said.
Boomi said its platform serves more than 30,000 customers and processes more than 200 billion messages a quarter. The company said the Agent Control Plane is part of the Boomi Enterprise Platform and is intended to help enterprises move AI deployments toward governed production use.
