BGL Democratizes Data Analysis for 200 Employees with Claude Agent SDK

by

The Era of Data Analysis for Non-Developers: Real-World Use Cases of the Claude Agent SDK

  • Australian financial firm BGL builds a text-to-SQL AI agent for all employees using the Claude Agent SDK
  • Secures security and scalability with Amazon Bedrock AgentCore, enabling 200 employees to analyze data without SQL
  • Key architecture: Data-driven separation + code execution pattern + modular knowledge structure

What Happened?

Australian financial software company BGL has built a company-wide BI (Business Intelligence) platform using the Claude Agent SDK and Amazon Bedrock AgentCore.[AWS ML Blog]

In simple terms, it’s a system where even employees who don’t know SQL can ask in natural language, “Show me the sales trend for this month,” and the AI automatically creates the query and even draws a chart.

BGL was already using Claude Code on a daily basis and realized that it was not just a simple coding tool, but had the ability to reason through complex problems, execute code, and interact autonomously with systems.[AWS ML Blog]

Why is it Important?

Personally, what makes this case interesting is that it shows a real answer to the question of “How do you deploy an AI agent into production?”

Most text-to-SQL demos work beautifully, but problems arise when you put them into real work. Table join errors, missing edge cases, incorrect aggregations. BGL solved this by separating the data foundation and the AI role.

They created well-refined analytical tables using existing Athena + dbt, and the AI agent is only focused on generating SELECT queries. Frankly, this is the key. If you leave everything to AI, hallucinations increase.

Another thing to note is the code execution pattern. Analysis queries return thousands of rows, sometimes megabytes of data. Putting all of this into the context window will cause it to crash. BGL had the AI directly execute Python to process CSVs from the file system.

What Will Happen in the Future?

BGL is planning to integrate AgentCore Memory. They plan to store user preferences and query patterns to create more personalized responses.

The direction this case shows is clear. By 2026, enterprise AI is evolving from “cool chatbots” to “agents that actually work.” The Claude Agent SDK + Amazon Bedrock AgentCore combination is one of those blueprints.

Frequently Asked Questions (FAQ)

Q: What exactly is the Claude Agent SDK?

A: It is an AI agent development tool created by Anthropic. It allows the Claude model to autonomously perform code execution, file manipulation, and system interaction, rather than just simple responses. BGL used it to handle text-to-SQL and Python data processing in a single agent.

Q: Why is Amazon Bedrock AgentCore needed?

A: Security isolation is essential for AI agents to execute arbitrary Python code. AgentCore provides a stateful execution environment that blocks access to data or credentials between sessions. It reduces the infrastructure concerns needed for production deployment.

Q: Is it actually effective?

A: 200 BGL employees are now able to perform analysis directly without the help of the data team. Product managers can test hypotheses, compliance teams can identify risk trends, and customer success teams can perform real-time analysis during client calls.

If this article was helpful, please subscribe to AI Digester.

References

Leave a Comment