Why vibe coding won’t cut it in capital markets

Vibe coding — the practice of prompting AI to generate software with minimal human oversight or structured engineering discipline — has captured the imagination in all industries. And it’s easy to see why. The idea that you can describe what you want and have working code appear in seconds is genuinely exciting. Collins Dictionary named “vibe coding” its 2025 Word of the Year, a reflection of how quickly the concept moved from developer shorthand to boardroom conversation. But in capital markets, excitement is not a compliance framework. And that, in a nutshell, is the problem.

The trend has reached financial services in a significant way. As early as June 2025, American Banker reported that employees at regulated banks were already vibe coding and that, while the practice has legitimate uses for prototyping and low-stakes tooling, the compliance and governance risks are real and largely unsolved. The piece captures a dynamic that will be familiar to anyone working in capital markets technology: the instinct to experiment is right, but the path from prototype to production is littered with obstacles that vibe coding does nothing to clear.

The scale of experimentation with vibe coding across industries puts that challenge in sharp relief: according to recent data, 63% of active vibe coding users today are non-developers — product managers, analysts and domain experts building software through natural language prompts alone. In financial services, where that population includes traders, operations staff and credit analysts who understand workflows intimately but have no obligation to understand secure software architecture, the implications are significant.
Anecdotally, many capital markets firms tell us that senior management is actively encouraging the use of AI throughout the enterprise, including experimentation by non-technical users.

The imperative is understandable. These are the people who understand the workflows, the data problems and the inefficiencies firsthand. But the code they produce through vibe coding almost always hits the same wall: IT can’t use it. It doesn’t meet security standards, it has no consistent architecture and it can’t be integrated into the firm’s technology ecosystem without essentially being rewritten from scratch. The result is wasted effort on both sides — business users who built something they can’t deploy, and IT teams who inherit a problem rather than a solution.

Financial markets are unforgiving environments. Applications must process millions of transactions with low latency and zero tolerance for error. They operate under layers of regulation — MiFID II, SEC, FINRA and more — where an audit trail is not a nice-to-have but a legal requirement. The stakes are not a broken webpage or a buggy mobile app. They are failed trades, regulatory penalties and reputational damage at institutions managing trillions of dollars.

Vibe coding, for all its speed, cannot account for any of this. AI models generating code from prompts have real limits in context, attention and control. They can produce something that looks right without being right. In a consumer app, you iterate and fix it. In a trading system, you don’t get that luxury.

There is also the question of what happens after the code is written. Capital markets software needs to be maintained, upgraded and audited over years, often decades. Code that was “vibed” into existence — with no consistent architecture, no standards-based foundation and no clear upgrade path — becomes a liability the moment it goes to production. Technical debt and shadow IT in financial markets is not abstract. It translates directly into operational risk and cost.

Gartner’s “Predicts 2026” report forecasts that software defects will increase by 2,500% by 2028, driven by business users creating tools inside the enterprise perimeter that security and IT teams have no visibility into, no ability to validate and no clear way to govern. Part of the spike is from a new class of defect where AI generates code that is syntactically correct but lacks awareness of broader system architecture and nuanced business rules. The remediation costs for these deep contextual bugs will consume budgets previously allocated to innovation.

Bringing the gap between AI-generated prototypes and production-ready software

What, then, is a viable alternative in capital markets? Bridging the gap between AI-generated prototypes and production-ready software requires a framework of guardrails, established architectural patterns, governance controls and human oversight. AI can accelerate application development, but it works best when operating within a framework that reflects the realities of regulated financial institutions.

Genesis provides this framework through its library of pre-built components engineered for the demands of capital markets and validated in production at the world’s largest financial institutions. These components provide the guardrails that let AI accelerate development without exposing firms to the risks that unconstrained AI coding creates. They also provide capital markets-grade security, compliance and performance, meaning that Genesis offers an efficient path from prototype to production because the chassis of each application is not generated by AI.

Critically, this approach also solves the business user problem. When Genesis offers AI in its application building tool, Genesis Create, it is not asking AI to improvise. With Genesis, AI assembles components and helps users build custom functionality by configuring specific endpoints built into the componentry. The process is fully transparent and gives the user ability to validate what AI is doing throughout the build. This unique combination of AI-powered assembly, industry-specific components and human oversight enables both businesspeople and professional engineers to achieve their goals.

They can even collaborate efficiently within the same project. A trader or analyst can use Create to define requirements, configure workflows and build out a working application, then hand it to an engineer for fine-tuning, integration and production hardening. The output at every stage is clean, maintainable and standards-based. It fits the firm’s existing architecture rather than fighting against it. For the first time, business users and IT are building toward the same goal rather than past each other.

Vibe coding offers a compelling shortcut. But in capital markets, shortcuts that bypass governance, compliance and architectural discipline are not shortcuts at all. They are detours that eventually lead somewhere expensive — and the wasted effort of business users building things IT cannot ship is fast becoming one of the industry’s most visible examples of that dynamic.

The future of AI in financial markets is not about removing discipline from software development. It is about making disciplined development dramatically faster. The firms that succeed will be those that combine AI’s speed with the governance, architectural standards and operational controls that regulated markets demand. Vibe coding may be an effective way to explore ideas, but production systems require something more durable: frameworks that enable innovation without sacrificing trust.