Agentic AI creates enterprise challenge beyond LLM boom

As organisations move beyond AI experimentation, a new challenge is emerging: how to connect, manage and govern potentially thousands of AI agents operating alongside human employees.

While much of the industry's recent focus has centred on LLMs and generative AI applications, attention is increasingly shifting towards the infrastructure required to deploy AI agents across enterprise environments.

Organisations are beginning to grapple with challenges that extend well beyond model selection, including onboarding AI agents, managing access to enterprise data and controlling operational costs.

One of the key issues facing enterprises is determining the appropriate level of reasoning capability required for different tasks, according to Kong Vice President and Field CTO, Solutions Engineering APJ, Ned Shawa.

"Not all tasks need the highest reasoning all the time," Shawa said.

"That burns tokens very quickly as well."

Kong is developing systems designed to match AI workloads with the most appropriate models based on task complexity, helping organisations optimise costs while maintaining consistency across deployments.

This approach aims to address a growing concern among enterprises experimenting with multiple AI models and agents: ensuring predictable outcomes and avoiding situations where different systems produce inconsistent responses for the same task.

"We're going to help with token optimisation and cost control, but most importantly, the repetition," Shawa explained.

"The minute you start getting different answers, you don't want that."

Building infrastructure for an agent-driven future

While enterprises are exploring agentic AI more and more, existing technology infrastructure was designed primarily for human users, not autonomous software agents.

Another obstacle facing the industry is the lack of established frameworks for integrating AI agents into organisational environments.

"This is a very green field," Shawa said.

"People are trying to figure out how to onboard agents into the organisation. 

"We figured out the human part of onboarding, but we never figured out how to onboard the agents."

The challenge extends beyond identity and access management. Modern enterprise environments typically contain data spread across numerous applications and platforms, creating silos that can be difficult for AI agents to navigate.

APIs were previously designed to enable software systems to communicate with one another. However, the emergence of agentic AI is creating demand for a new generation of infrastructure specifically designed for autonomous agents.

To address this, Kong is investing in AI connectivity and agentic infrastructure, built around APIs and emerging standards such as the Model Context Protocol (MCP).

Agents require direct, structured access to enterprise systems and data sources, and experts increasingly view MCP as a potential standard for enabling secure communication between AI systems and enterprise applications.

More than API management

The emergence of AI connectivity has prompted questions about whether the sector represents a genuinely new technology category or simply a rebranding of traditional API management.

But the foundations have been in place for years.

The company originated in the API space, with founders envisioning APIs as the primary mechanism for digital communication. The rise of AI agents is now accelerating that vision, transforming APIs from developer-focused tools into critical infrastructure for machine-to-machine interactions.

As a result, traditional APIs are evolving to accommodate the requirements of AI systems, with MCP increasingly viewed as a complementary framework that enables agents to discover, access and interact with enterprise services more effectively.

It reflects a broader industry trend as software vendors, cloud providers and enterprise technology companies race to establish standards for agent interoperability.

Agent governance a key priority

Looking into the near future, governance and registration of AI agents is going to become increasingly important.

Kong is exploring future capabilities centred on agent registration and lifecycle management.

Such capabilities could become critical as organisations deploy large numbers of autonomous agents across departments and business functions.

Enterprises may soon face challenges similar to those experienced during the rapid growth of cloud services and SaaS platforms, where visibility and governance often lagged behind adoption.

For AI agents, those challenges could be amplified given their ability to make decisions, access information and interact with business systems autonomously.

APAC regulation still varies

Regulatory approaches remain unaligned across APAC, with different markets advancing AI governance at varying speeds.

Australia, New Zealand and Singapore are among the markets actively working to establish frameworks around AI accountability and decision-making.

As agents become capable of making operational decisions, organisations will need greater transparency into how those decisions are reached.

"If we now have agents making decisions, we should, at least, have an understanding of how the decision was made," Shawa said.

The differing approaches create both opportunities and challenges for technology providers seeking to deploy AI infrastructure across multiple jurisdictions.

From education to implementation

While enthusiasm surrounding AI remains high, many organisations are still in an education phase rather than full-scale deployment.

CISOs and CTOs continue to seek guidance on implementation strategies, governance frameworks and operational blueprints for agentic AI.

That phase is likely to accelerate over the coming months, with enterprises shifting from experimentation to implementation in 2027.

There is still a lack of established best practices for managing large-scale agent deployments. That becomes increasingly significant when considering the scale organisations may eventually need to support.

Most employees could eventually work alongside multiple AI agents, effectively creating digital teams that augment human productivity.

"When we have a new agent coming in, every person will end up with three or five agents working beside them," Shawa said.

For large enterprises employing thousands of workers, that could translate into tens of thousands of AI agents operating simultaneously.

Software engineering teams may be among the earliest adopters, with agents acting as assistants that help developers become significantly more productive.

The scale of that transformation remains uncertain, but it's quite possible that organisations are currently underestimating how rapidly agent adoption could occur.