AI Coding in Fintech: From Code Gen to Full SDLC Assistance

The New Frontier: AI Beyond Model Power

Large language models keep getting smarter, but the next phase of competition isn't just about the models themselves. It's about building reliable agents, crafting solid AI engineering systems, and making AI work in messy, real-world business environments. That's the shift we're seeing across the industry.

On August 21–22, AICon Global AI Development and Application Conference lands in Shenzhen. The full agenda is live, and it's packed with technical talks from people who build AI systems for a living. Topics range from agent engineering and large model infrastructure to AI-native development and embodied intelligence. The throughline: how do we move AI from flashy capability to dependable system?

HSBC's Take: Open Source Inside

One session that caught my eye comes from Li Weining, HSBC Technology's internal open source lead. He's speaking in the track called "AI Native New Paradigm: Coding Agents Rebuilding the Software Development Lifecycle." His talk: "From Code Generation to Development Loop: AI Coding in Fintech SDLC."

Li has spent 15 years in fintech, wearing hats in R&D, testing, ops, architecture, delivery, and product management. He's worked at HSBC, GAC Auto Finance, Xiaopeng Financing Lease, and Hang Seng Bank. So when he talks about AI coding in regulated environments, he's not theorizing.

Why Fintech Is a Special Beast

Fintech software development isn't like building a consumer app. You're dealing with strict security, compliance, and risk requirements. A hallucinated code snippet that works in a demo could be a disaster in a trading system or a loan approval pipeline. That's why HSBC's approach to AI coding isn't just about generating code faster—it's about embedding AI into the entire software development lifecycle, from requirements to delivery.

The key pain points Li identifies are familiar: inconsistent model output, security and compliance risks, and the difficulty of scaling AI tools across teams with different tech stacks and workflows. But the bigger challenge is cultural. You can't just hand developers a shiny new tool and expect them to adopt it. You need to shift the platform, the processes, and the mindset.

Internal Open Source: A Community of Practice

HSBC's answer? Internal open source. Instead of a top-down mandate, they're pooling the AI coding experiences of different teams. Developers share what works, what doesn't, and what they've learned. This isn't just about collecting prompts—it's about turning scattered know-how into reusable tools and "Agent Skills."

Think of it as an internal GitHub for AI-assisted development practices. Teams contribute their successful patterns, and others can adopt them. This community-driven approach builds a library of vetted, governance-ready components that anyone can use. It's a smart way to avoid reinventing the wheel and to ensure that AI adoption doesn't happen in silos.

Agent Skills Across the SDLC

Li's talk walks through specific use cases for Agent Skills at each stage of the SDLC. Let me break down what that looks like in practice.

Requirements Phase

AI agents can help parse Jira tickets and Confluence pages to build context. They can ask clarifying questions, flag ambiguities, and even suggest acceptance criteria. This isn't about replacing the business analyst—it's about giving them a superpowered assistant that never forgets a detail.

Design Phase

When a developer is about to architect a new module, an agent can generate options, perform impact analysis, and lay out trade-offs. It's like having a senior architect on call 24/7. But the human still makes the call—the agent just makes sure you've considered the angles.

Coding Phase

This is where most AI coding tools start: in the IDE. HSBC integrates with VS Code and GitHub Copilot to help developers write code faster. But the real value comes from the context. The agent knows the project's conventions, the relevant libraries, and the patterns that have worked before.

Code Review

AI-assisted code review is a different beast. It's not just about catching bugs—it's about flagging security risks, checking compliance with internal standards, and identifying potential architectural issues. The agent acts as a second pair of eyes that never gets tired.

Testing

Generating test cases is a natural fit for LLMs. But the harder part is closing the loop: running the tests, analyzing failures, and suggesting fixes. HSBC's agents aim to do just that, turning testing from a chore into a continuous feedback cycle.

MCP: The Glue for Enterprise Workflows

A big part of making agents useful is getting them into the flow of work. That's where the Model Context Protocol (MCP) comes in. MCP is an open standard that lets AI models connect to external tools and data sources. In HSBC's case, it's what ties together VS Code, GitHub Copilot, Jira, Confluence, and other enterprise systems.

Instead of a bunch of point-to-point integrations, MCP provides a unified way for agents to access the tools and data they need. This is crucial in a financial institution where data sits in multiple systems, each with its own access controls. MCP lets agents pull in the right context without exposing sensitive information.

Governance and Security: The Non-Negotiables

You can't just let AI agents run wild in a bank. Li's talk dedicates a whole section to security, compliance, and risk management. That means thinking about data boundaries, access control, and audit trails. Agent Skills need to be versioned, permissioned, and reviewed just like code.

There's also the question of balancing efficiency with risk. AI can speed things up, but if it introduces errors or security holes, the cost could be huge. HSBC's approach is to make governance part of the AI tooling itself, not an afterthought.

Scaling from Pilot to Enterprise

The final piece is scaling. You don't go from a pilot project to a company-wide platform overnight. HSBC started with small internal open source projects and gradually grew to a platform that thousands of engineers use and contribute to.

The trick is to pick high-value pilot scenarios that are repeatable. Once you have a few wins, you can build a playbook for others. And you need to think about developer adoption—training, feedback loops, and community management. It's not just a tech rollout; it's a culture change.

What This Means for Assistive Technology

Here's the thing: this talk is squarely about assistive technology. AI coding tools are assistive tech for software engineers. They help people do their jobs better, not replace them. And the lessons HSBC is learning—about community, governance, and scaling—apply to any assistive technology deployment.

Whether you're building screen readers for visually impaired users or AI copilots for developers, the core challenges are the same: How do you make the tool reliable? How do you keep it safe? How do you get people to actually use it?

HSBC's internal open source model is a compelling answer. By treating AI coding practices as a shared resource, they're building a culture of continuous improvement. And that's something any assistive tech team can learn from.

Key Takeaways

• AI coding is evolving from a personal productivity hack to a full SDLC capability.
• Internal open source and community co-creation can accelerate adoption and ensure quality.
• Agent Skills can be applied at every stage—requirements, design, coding, review, testing, and delivery.
• MCP and tool integration are critical for making agents work within existing workflows.
• Security, compliance, and governance must be baked into the AI tooling from day one.
• Scaling requires a thoughtful approach: pick high-value pilots, build a playbook, and invest in developer community.

If you're in Shenzhen in late August, this is a session worth catching. And if you're not, the takeaways are still valuable. AI is becoming a part of how we work, and learning from pioneers like HSBC can help all of us build better assistive tools.