Asia and the Pacific has entered a new phase of trade digitalization. Many customs and border agencies already use decision-support AI for document extraction, risk management, cargo targeting or classification support, as revealed by the upcoming results of the ESCAP-ADB Survey on AI in Trade Facilitation. As AI use and capabilities expand, the question now is whether governments should move from AI tools that analyse and recommend to AI agents that can act and help coordinate multi-step trade workflows. That question is especially timely because digital trade systems are expanding while governance and preparedness still lag technological capability, a gap that has potentially important sovereignty implications in the context of public services provision.

The distinction matters. Decision-support AI helps officials classify, extract, score, explain or recommend. Generative AI can add natural-language functions such as summarizing files, answering trader questions or drafting text. But AI agents go further: they use models together with memory, planning, rules and tools to move a task toward completion. In other words, an AI agent helps execute a workflow and may be designed to make significant decisions and take concrete actions.

For trade facilitation, the attraction of AI agents is obvious. Once documents and data become digital, the main bottlenecks are no longer scanning or filing, but coordination across customs, port, sanitary and phytosanitary, licensing and transport systems. A bounded agent could assemble a draft declaration, identify missing supporting documents, route a case across a single window, reconcile status updates, or prepare an audit-ready file for officer approval – or approve it autonomously if so designed. These are precisely the kinds of repetitive, cross-system administrative tasks that delay clearance without necessarily requiring legal discretion. The World Customs Organization (WCO) and International Monetary Fund (IMF) guidance also explicitly links reducing human interaction at borders with lower corruption risks, including through automation and the use of agentic AI for risk analysis.

Yet this is also where caution is needed. ESCAP’s work on cross-border paperless trade shows that digital trade facilitation succeeds only when legal readiness, technical interoperability and institutional governance are in place. The Cross-border Paperless Trade Toolkit and the Framework Agreement on Facilitation of Cross-border Paperless Trade in Asia and the Pacific (CPTA) both point in the same direction: before countries automate across borders, they need trusted data exchange, harmonized processes and a clear governance framework. Without those foundations, agents risk automating fragmentation rather than facilitation.

Emerging examples are promising but still limited. A recent proof of concept by Microsoft with ANZ, HSBC and Lloyds showed how AI agents embedded in enterprise systems could extract, validate and transmit structured trade data in a standards-based workflow. In Europe, early agentic customs and trade-compliance platforms are starting to automate document checking, discrepancy detection and draft submission preparation (e.g., UMG AG Customs autopilot). Live deployments, in particular across government agencies, are still extremely rare, with AI agents mostly limited to generative AI chatbots to provide information or collect feedback from users, e.g., in Dubai or in the United States. Still, the number of pilots is growing and examples suggest that agentic trade workflows are becoming technically feasible. They also confirm that the field is still at an early stage and far from routine public-sector deployment.

As AI systems become more autonomous, the governance challenge changes. The issue is no longer simply whether AI can improve efficiency, but whether automated actions remain transparent, reviewable and accountable.  In trade facilitation, where administrative actions can affect compliance outcomes and border decisions, AI agents must operate within clearly defined procedural and institutional safeguards. This will also drive the choice of AI ecosystems between on premises open-source models and potential external private providers operating in a cloud.

So, what should governments and trade-related agencies do now?

First, keep investing in decision-support AI where value is already proven: document processing, explainable risk scoring, translation, classification support and trader guidance. 

Second, pilot AI agents only in low-risk administrative workflows where actions are reversible and human approval remains mandatory. 

Third, strengthen the policy guardrails that matter most: legal authority, auditability, cybersecurity, and environmental sustainability. That last point matters more than before. ESCAP has shown that trade digitalization can reduce waste and emissions, while the latest Stanford AI Index report warns that AI’s environmental footprint is rising fast. The right approach is therefore to prioritize AI solutions that will use energy and natural resources in a sustainable manner.

Progress in AI applications over the past few years has been incredibly rapid, and it seems plausible that AI agents will become ubiquitous over the next decade. The potential for trade facilitation, reducing trade costs, fighting corruption and increasing transparency is clear. The challenge will be to define clear boundaries and develop robust oversights for these AI agents, ensuring that “Human in the Loop” mechanisms consider both agent reliability and system integrity. 

Careful testing and gradual integration of AI agents is the way forward. In Asia and the Pacific, the neutral and enabling platform provided by the CPTA and ESCAP may be leveraged to exchange practices and lessons learned in this area, and develop safe and secure interoperable automation solutions.