AI’s Current Limits in Rebuilding Complex Financial Tools, According to Linas Beliūnas

In a recent LinkedIn post, Linas Beliūnas discusses the current capabilities and limitations of artificial intelligence in recreating sophisticated financial software, using the example of an attempt to rebuild the Bloomberg Terminal.

Beliūnas shared his experience using Perplexity AI to generate code for a Bloomberg Terminal replacement. He detailed the impressive scope of the AI’s output, noting the inclusion of features such as:

“Real-time market dashboards
Equity & FX charts with technical overlays
Earnings summaries
Macro calendar
Company profiles
AI-generated research briefs
News feed with sentiment tags
A dark mode UI that looked… expensive”

Despite the AI’s ability to generate a substantial amount of code—reportedly 80,000 lines in three hours—Beliūnas highlights that the resulting application was non-functional. He pointed out critical issues that prevented the software from working correctly.

Data Normalization and Integration Challenges

According to Linas Beliūnas, a primary hurdle was the lack of normalized data sources. This means that the data from various financial markets and instruments was not in a consistent or compatible format, making it difficult for the application to process accurately.

Beliūnas further elaborated on the technical difficulties, stating:

“Data sources weren’t normalized.
Tickers mismatched across exchanges.
Some numbers were… creative, to say the least.”

This suggests that while AI can generate code and mimic interfaces, it currently struggles with the intricate data wrangling and validation required for high-stakes financial applications where precision is paramount.

The Gap Between Aesthetics and Functionality

An interesting observation from Beliūnas’s post was the AI’s success in creating an aesthetically pleasing user interface, even if the underlying functionality was absent. He humorously noted, “But boy, was that dark mode beautiful.” This highlights a common trend where AI can excel at design elements and front-end presentation, but falters when faced with the complex, real-world data integration and logical processing needed for robust applications.

AI’s Role in Startup Operations

While Beliūnas expressed skepticism about AI’s immediate ability to replace complex financial terminals, he also pointed to its potential in other areas. In a postscript, he directed readers to a separate resource where he details how he built an AI operating system to run a startup using Claude.

In Linas Beliūnas’s view, this demonstrates that while AI may not yet be ready to replicate the entirety of a Bloomberg Terminal, it can be a powerful tool for streamlining specific business operations and potentially enabling lean startup models. His experience underscores the nuanced understanding required to leverage AI effectively, recognizing its current strengths in areas like code generation and design, while acknowledging its weaknesses in complex data integration and functional reliability for highly specialized software.