In a recent LinkedIn post, Linas Beliūnas offers a stark, and somewhat humorous, account of attempting to leverage advanced AI for a critical fintech application. Beliūnas shares his experience asking Claude AI to generate code for a complete digital banking stack, akin to a Revolut replacement. The exercise, which took approximately five hours and resulted in 80,000 lines of code, aimed to build out core functionalities including KYC onboarding, account creation, virtual cards, transfers, bill pay, subscription management, fraud alerts, admin dashboards, and even auto-generated compliance documents.
Beliūnas highlights the ambition of the AI’s output, noting:
“It even had confetti animation when the balance went up! 🎉”
However, the post quickly pivots from the impressive-sounding output to the harsh reality of its practical application. According to Beliūnas, the generated code was fundamentally flawed, leading to a cascade of critical system failures.
The Chasm Between AI Code Generation and Functional Fintech
Linas Beliūnas emphasizes that despite the volume and apparent complexity of the AI-generated code, it failed to deliver a working product. He details the severe consequences encountered:
“None of it worked. Transactions failed. Balances desynced. Customers couldn’t pay. We were in full panic mode, and regulators were calling nonstop.”
This experience, as Beliūnas recounts, underscores a significant gap between AI’s capability to produce code and its ability to create robust, reliable, and compliant financial systems. The anecdote serves as a cautionary tale for businesses considering full AI integration for mission-critical infrastructure without rigorous human oversight and testing.
The Illusion of AI-Driven Innovation
Beliūnas’s post subtly critiques the hype surrounding AI’s current capabilities in complex software development. While the AI managed to replicate the *structure* of a digital banking platform, it lacked the understanding and precision required for the intricate, interconnected logic that underpins financial transactions. The seemingly trivial detail of a confetti animation, while a testament to the AI’s ability to mimic user interface elements, stands in stark contrast to the core banking functions that failed spectacularly.
As Beliūnas wryly concludes, in reference to the confetti animation:
“But boy, was that confetti beautiful.”
This juxtaposition highlights the potential for AI to create impressive-looking, yet ultimately non-functional, solutions when applied to domains demanding absolute accuracy and reliability.
Serious Implications and the Future of AI in Startups
Beyond the immediate technical failures, Beliūnas points to the serious business and regulatory implications that arise from such a situation. The mention of regulators calling nonstop indicates the high stakes involved in financial services, where errors can have significant legal and financial repercussions.
On a more serious note, Beliūnas directs readers to another resource he has developed, suggesting a continued exploration of AI’s practical applications. He shares a link to his work on building an “AI operating system to run a startup with Claude,” indicating that while his experiment with a full banking stack revealed limitations, he remains engaged with leveraging AI for business efficiencies. This suggests a nuanced view: AI can be a powerful tool, but its implementation requires careful consideration of its current limitations, especially in highly regulated and complex industries like fintech.
📝 About This Content
This article is based on insights shared by Linas Beliūnas on LinkedIn.
📅 Originally posted on February 20, 2026 | View original post on LinkedIn →