RECENT advances in large language models and agentic architectures have fundamentally transformed artificial intelligence (AI) capabilities. Today’s AI systems can plan multi-step tasks, reflect on their outputs, use tools and even coordinate with other AI systems.
We are witnessing the emergence of truly agentic AI – systems that operate with increasing autonomy and goal-directed behaviour rather than merely responding to prompts. However, as these impressive capabilities mature, a critical infrastructure gap threatens to undermine their potential at scale.
The problem is straightforward but profound: most AI agents today are confined to proprietary technological stacks. They rely on platform-specific memory stores, orchestration logic, toolchains and interaction schemas that work perfectly well in isolated environments, but create serious friction when deployed across organisational boundaries.
Consider what happens when a business process spans multiple systems, from customer interaction to order fulfilment, or across supply chain and finance departments. Agents built by different vendors cannot share crucial context, delegate subtasks or coordinate actions effectively. As a result, cross-domain AI-driven automation, which should be the primary value proposition of agentic systems, becomes unworkable in practice.
BT in your inbox
Start and end each day with the latest news stories and analyses delivered straight to your inbox.
AI agents must interoperate across nine dimensions
Solving agent interoperability requires addressing multiple interconnected layers, not just creating superficial application programming interface (API) connections. If agents are to coordinate effectively across organisational, technical and vendor boundaries, we need a common substrate for how they discover each other, delegate tasks, share knowledge and enforce constraints.
Tool use represents the most visible dimension, where agents must invoke external services, automation platforms and APIs without relying on brittle hard-coded logic.
Anthropic’s Model Context Protocol, for instance, tackles certain aspects of tool access. Communication protocols are equally critical, as agents need structured dialogue mechanisms to assign roles, exchange information, and negotiate task handoffs between systems. Cisco and LangChain’s AGNTCY standard and Google’s brand new A2A framework provide initial approaches to how agents may talk to each other.
But the requirements quickly expand beyond these basics. Trust and security become foundational concerns in multi-agent environments, requiring shared models for identity verification and authorisation. Memory systems must provide persistent, cross-context storage that agents can access across organisational boundaries.
Knowledge sharing demands mechanisms not just for information access, but for verification and integration of information across different agent systems. Looking toward more advanced interoperability, transaction capabilities become essential for agents to negotiate, purchase and compensate across digital marketplaces.
Stripe has developed preliminary agent toolkits for economic exchange. Governance frameworks must move beyond static policies to become executable constraints embedded directly in agent workflows. Discovery protocols are needed so agents can dynamically locate and interact with other relevant agents in a secure way, without relying solely on preconfigured integrations.
Finally, distributed agent systems require standardised approaches to error handling, where failures can be systematically surfaced, escalated or resolved.
These are some of the promising early efforts addressing pieces of this puzzle. However, they remain point solutions in different stages of maturity, enterprise readiness and adoption levels, indicating that a more comprehensive and coordinated approach is needed for true interoperability.
AI agents are path forward for global innovation
Interoperability should not be viewed as the enemy of commercial differentiation but rather, as its enabler. Open ecosystems do not diminish commercial opportunity; they expand it dramatically. When cloud platforms adopted container standards and interoperable storage APIs, they did not sacrifice competitive advantage; instead, they created a larger surface area for value-added innovation. The same principle applies to agentic AI systems.
Vendors supporting interoperable agent frameworks can still compete vigorously on orchestration strategies, domain specialisation (for example, offering agents pre-trained for specific industries like healthcare or finance), tooling sophistication and developer experience.
But they compete within a viable, expanding ecosystem rather than through isolation tactics that ultimately limit total market growth. For enterprise customers, the advantages of interoperability are even more compelling, enabling automation that spans organisational boundaries without requiring wholesale commitment to a single vendor’s vision.
Asia-Pacific could be a key proving ground for agentic AI
With its unique combination of cross-border trade complexity, regulatory diversity and vibrant consumer services sector, Asia-Pacific creates real-world conditions that demand adaptive, autonomous systems. This unique combination suggests that the Asia-Pacific region could serve as a critical proving ground for these interoperable agentic technologies.
Agentic AI will find some of its most valuable use cases in environments where integration across languages, infrastructure types and regulatory jurisdictions is not optional but essential for business operations.
Achieving true interoperability will require more than individual product road maps or incremental feature development. It necessitates a coalition approach spanning technology vendors, industry associations, standards bodies, regulatory agencies and enterprise technology buyers. Interoperability must be treated as strategic infrastructure, not relegated to an afterthought or post-implementation integration problem.
The fundamental choice facing the AI industry is not between control and openness. It is between creating composable, scalable intelligence versus isolated automation silos. Those who solve for the complex interoperability requirements of diverse markets like Asia-Pacific will likely establish the blueprint for agentic ecosystems globally. And the time to make these architectural decisions is now, before proprietary approaches calcify into barriers we cannot undo.
The writer is a principal analyst at Forrester
