How AI Coding Is Reshaping Import-Export Software Development

The Software Bottleneck in Global Trade

If you run an import-export business, you know the pain of paperwork, customs delays, and regulatory shifts. But there's a quieter bottleneck—the software that runs your operations. Legacy systems are brittle, updates are slow, and developers are scarce. That's where AI coding comes in. It's not about replacing programmers; it's about making them faster and more accurate.

At the recent AICon Shenzhen conference, Li Weining, head of internal open source at HSBC Technology, laid out how AI coding is transforming the software development lifecycle (SDLC) in fintech. While his talk focused on banking, the lessons apply directly to trade finance, logistics, and customs software. Import-export companies face similar challenges: complex regulations, high-stakes data, and the need for rapid adaptation.

From Code Generation to Full-Cycle Development

AI coding started as a fancy autocomplete. Tools like GitHub Copilot could suggest snippets, but they didn't understand the bigger picture. Li's talk emphasized a shift: from generating code to automating the entire SDLC. That means requirements analysis, architecture design, code review, testing, and deployment—all touched by AI.

For import-export, this is huge. Imagine an AI that reads a new customs regulation and automatically updates your compliance checks, then suggests code changes to your shipping module. That's the vision. The tech is already there; the challenge is integration.

Why Internal Open Source Is the Secret Sauce

HSBC's approach is built on internal open source. Teams across the company share their AI coding experiments, prompts, and tools. Instead of reinventing the wheel, they build on each other's work. This isn't just a nice-to-have—it's essential for scale.

In trade finance, different teams handle letters of credit, bills of lading, or customs declarations. Each has unique quirks. If one team discovers a clever way to use AI for trade document verification, others can adopt it quickly. Internal open source turns isolated wins into company-wide capabilities.

Agent Skills: The Building Blocks of Smart Automation

Li introduced the concept of 'Agent Skills'—reusable AI capabilities that can be plugged into different parts of the workflow. These aren't just prompts; they're structured tools with governance, permissions, and audit trails. In the import-export world, you could have an Agent Skill for Incoterms analysis, another for tariff classification, and one for export control screening.

These skills work inside your existing tools. HSBC integrates with VS Code, GitHub Copilot, Jira, and Confluence. For a logistics software team, that means AI assistance right where you're already working. You don't need a separate AI platform; you need AI that lives in your development environment.

Practical Use Cases Across the SDLC

Let's get concrete. In the requirements phase, an AI agent can pull up relevant Jira tickets and Confluence pages, summarize what's needed, and ask clarifying questions. In design, it can generate architecture options and flag risks. In coding, it suggests implementations based on your codebase's patterns. In review, it spots potential bugs and compliance issues. In testing, it generates test cases and even helps debug.

For an import-export system, imagine a new feature to calculate duties based on a trade agreement. The AI can read the agreement, propose the logic, generate tests for different scenarios, and review the code for compliance. That's a massive time saver.

Real Pain Points: Stability and Security

But it's not all rosy. Li highlighted three big pain points. First, AI outputs can be unstable—hallucinations are real. In a customs context, a wrong HS code could cost thousands. Second, security and compliance are non-negotiable. AI might leak sensitive data or call tools without proper authorization. Third, scaling across teams is hard. Different tech stacks and processes make it tough to standardize.

These are the same issues import-export software teams will face. The key is to start small, pick high-value use cases, and build governance early.

Tool Integration: Making AI Work for You

One of the most promising developments is MCP (Model Context Protocol), which connects AI to your tools and data. Li showed how MCP lets agents pull real-time data from Jira, Confluence, and even code repositories. That's powerful for trade systems that rely on live shipment tracking or customs updates.

But integration isn't just about tech—it's about workflow. You need to design how agents interact with humans. For instance, an AI might draft an email to a customs broker, but a human must approve it. The goal is to reduce friction, not add more.

Governance: The Non-Negotiable for Trade Compliance

In financial services, and equally in import-export, governance is everything. Li stressed the need for AI skills to have permissions, audit logs, and quality checks. You can't have an AI making unauthorized calls to a customs API or accessing sensitive commercial data.

For import-export companies, this means setting clear boundaries. Define what AI can and cannot do. Ensure every AI action is logged. Regularly review AI-generated code for compliance. It's a lot of work, but it's the only way to scale safely.

Scaling from Pilot to Enterprise

HSBC started with pilot projects and grew to a platform used by thousands. The secret? Choosing high-value scenarios, measuring results, and building a community of practitioners. They also focused on developer experience—if developers don't like the tool, they won't use it.

For an import-export firm, start with a single team, say the trade finance unit. Identify a repetitive task that AI can speed up. Track metrics like time saved and error rates. Then expand. Before you know it, AI coding becomes part of your DNA.

The future of import-export software isn't just about better algorithms—it's about better workflows. AI coding is the vehicle. The question is: are you ready to ride?