AI Coding in FinTech SDLC: A Health Informatics Perspective

From Code Generation to Full-Stack Development: The AI Shift

The AI conversation has moved past model benchmarks. Now it's about building reliable agents and weaving them into complex workflows. This is especially true in regulated industries like finance and healthcare, where getting it wrong isn't just a bug—it's a compliance nightmare.

At the recent AICon Shenzhen conference, HSBC's internal open-source lead, Li Weining, laid out a practical roadmap for taking AI coding from a personal productivity hack to an organizational capability. While his talk focused on fintech, the parallels to health informatics are striking. Both sectors wrestle with legacy systems, strict data privacy rules, and a culture that rewards caution over speed.

The Real Pain Points in FinTech—and Why Health IT Should Care

Li didn't sugarcoat the challenges. AI coding tools still hallucinate, produce inconsistent code, and raise serious security flags. In fintech, a single misstep could leak sensitive data or trigger unauthorized actions. Health informatics faces the same fears: patient records, protected health information, and the FDA's watchful eye.

He also noted that rolling these tools out across teams is tough. Different groups use different stacks and workflows, and proving the ROI is a moving target. It's not just a tech problem—it's a people and process problem.

Building a Community of Practice Through Inner Source

HSBC's answer? Treat AI coding practices like open-source software. They created an internal open-source program where teams share their experiences, prompts, and tools. Instead of everyone reinventing the wheel, they pool their insights into reusable Agent Skills.

This isn't just about being nice. It's about governance. By curating what works, they can standardize, audit, and ensure compliance. For health informatics, this model could be a lifeline. Imagine a shared library of validated AI prompts for clinical documentation or automated test generation—vetted for safety and interoperability.

Agent Skills Across the SDLC: A Stage-by-Stage Look

Li walked through the entire software development lifecycle (SDLC), showing how Agent Skills can assist at each step. It's not just about writing code faster.

Requirements: Making Sense of Jira and Confluence

In the requirements phase, agents can help parse user stories, clarify ambiguity, and connect the dots across project management tools like Jira and Confluence. For health IT, this could mean translating clinical workflows into technical specs with fewer misunderstandings.

Design: Generating and Evaluating Architecture

For design, agents can suggest architecture options, run impact analyses, and even weigh trade-offs. In healthcare, where systems need to handle sensitive data and integrate with electronic health records, these insights could save months of back-and-forth.

Coding: Boosting Developer Efficiency

During coding, tools like GitHub Copilot and VS Code integrations help developers write boilerplate and tests faster. Li emphasized that the real win isn't just speed—it's consistency and adherence to internal standards.

Review: Catching Risks and Violations

Code review gets a boost too. Agents can flag potential security issues, style violations, and even logic errors before a human reviewer steps in. In a field like health informatics, where audit trails matter, this is gold.

Testing: Generating and Closing the Loop

Testing is where agents shine—they can generate test cases, analyze failures, and help close the verification loop. Li's team uses this to ensure quality without burning out developers.

Tool Integration: Making Agents Part of the Workflow

None of this works in a vacuum. HSBC leverages MCP (Model Context Protocol) to connect agents to the tools and data they need. It's about moving from standalone assistants to a cohesive agent workflow.

For health informatics, imagine an agent that pulls patient de-identification rules from a policy database, checks them against FHIR standards, and then helps write code that complies with both. That's the kind of integration that could make AI coding viable in clinical settings.

Governance, Security, and the Path to Scale

The elephant in the room is governance. Li stressed that in fintech, you can't just let AI run wild. You need clear boundaries, audit trails, and risk assessment processes. Agent Skills must be versioned, permissioned, and reviewed.

He also talked about scaling—moving from a few pilot teams to thousands of developers. The key is picking high-value use cases, building a feedback loop, and fostering a culture that embraces change. For health informatics, this means starting with low-risk admin tasks, proving value, and then expanding to more critical applications.

What Health Informatics Can Borrow from FinTech's Playbook

Li's talk was a masterclass in pragmatism. He didn't promise magic. He offered a systematic approach to AI adoption in a risk-averse industry. Health informatics faces similar constraints—maybe even stricter ones.

The inner-source model is a standout idea. It turns AI coding from a solo sport into a team effort, with shared learning and built-in governance. The stage-by-stage application of Agent Skills shows that AI isn't just for the coding phase; it can enhance the entire SDLC. And the emphasis on tool integration via MCP points to a future where AI agents are embedded in the very fabric of development workflows.

If you're in health informatics, don't wait for the perfect AI. Start small, share what you learn, and build a community of practice. That's how you turn a shiny tool into a reliable partner—without sacrificing safety or compliance.