How Our Engineering Team Embraced AI and Claude Code for 10x Productivity

Treasure Data’s platform enables the world’s largest companies to assemble and activate customer data and AI agents they can trust. But we’re not just empowering our customers to do more with AI – we’re also introducing more intelligent workflows across our organization to accelerate work. 

Treasure Data’s engineering team is one such team embracing AI. The team is on a mission to achieve 10x productivity through the use of AI and agentic coding. In 2025, they have embraced Claude Code, an agentic AI tool that’s reshaping how the team is designing, writing, and deploying code. 

A proactive approach to testing and evaluating AI tools has led to Claude Code and other AI tools becoming a fundamental part of the development workflow, accelerating project timelines and task completion. It’s even expanding into other departments, such as customer support. 

Read on to learn about Treasure Data’s fast-paced journey with new AI coding tools and how Claude Code went from a promising new tool to a trusted engineering partner.

Adopting Claude Code in engineering workflows

To understand how Claude Code has improved engineering work, it’s important to understand how the team worked before they started using AI coding tools.

First, writing code is only part of the work engineers do. The development process typically involves several steps: planning or learning new technologies, writing code, testing, and deploying. This process typically happened in two-week sprints.

The engineering team started using GitHub Copilot for inline code completion in the coding editor and Google Gemini for other coding tasks. Over time, approximately 70% of Treasure Data’s engineers used AI-assisted coding tools. In the spring, Copilot expanded its capabilities with a chat interface that can generate code or test code for engineers. Engineers used the new coding interface, but they still had to write a prompt every time they wanted new code, then accept the code and ask Copilot to perform the next task. While it helped them, it was still time-consuming.

Claude Code changed things. Claude Code is designed to work seamlessly within the command-line interface and includes the ability to utilize advanced terminal tools like grep to locate code snippets and analyze project files. This approach eliminates the need to copy and paste code files and projects.

With Claude Code, the entire process is sped up. An engineer can provide the internal knowledge of how the Treasure Data system works and the preferred technology to use, creating an initial design note for the AI. Claude Code then enhances that description to create a proof-based implementation design and starts writing the code, while the engineer reviews the code changes.

Because Claude Code is integrated into the machine’s terminal, Saito said they were at first concerned about letting it run commands. But once they saw what became possible, they realized how useful that was. Unlike fully autonomous coding tools, Claude Code breaks tasks into logical steps, using prompts and confirmation requests to give engineers full visibility over what the AI agent is doing. This ensures code quality, eliminates surprises, and helps engineers learn from each step.

The engineering team hasn’t abandoned using Copilot. Its use is evolving for different use cases, including using it when using the code editor.

Redefining how code gets built with agentic coding

Today, agentic coding is a huge part of the engineering team’s process. Agents are performing whole engineering processes. They are not only writing code, but building, testing, and deploying it. For example, an engineer explains their desired features or changes (e.g., add a “New Chat” button or replace a sidebar with chat history) as a prompt, and Claude Code performs the following actions:

•     Analyzes the project structure and existing codebase.
•     Writes new code as required and tests it within the terminal.
•     Fixes bugs, builds the project, and ensures functionality before committing changes.

Helping every team member, from interns to principal engineers

Agentic coding has proven beneficial for engineers across all experience levels.

For Sean Isayama, Engineering Intern, using Claude Code has made his internship easier by streamlining his work on assigned tickets and bugs, and giving him more time to understand Treasure Data’s specific APIs and technologies. Isayama said he can give Claude Code a general task, and the agent will perform all the stages in agentic coding for him. 

Even when tasked to work on a repository in an unfamiliar coding language, he was able to quickly understand the codebase through a series of quick conversations with Claude Code, and was ready to take ownership of his first feature.

Experienced Treasure Data engineers can also now offload some of their work to AI, allowing them to focus on company-specific aspects of the coding process. Before using Claude Code, Taro Saito, Senior Principal Software Engineer, had to dedicate 100% of his time to writing code. Now, he can perform other engineering tasks and provide them as inputs to the AI, allowing for parallel task work.

Making 10x productivity a reality: Agentic coding in action

The best example of how Claude Code has improved the engineering team’s work is the Treasure Data MCP Server. An MCP (Model Context Protocol) Server acts as a secure, real-time communication channel between AI assistants and the tools and data they need access to.

The Treasure Data MCP Server is an open-source tool that connects Large Language Model (LLM) assistants like Claude, GitHub Copilot Chat, and Windsurf directly to the Treasure Data environment, enabling users to explore databases, analyze tables, and write SQL queries using nothing more than plain English.

While this project would normally take 2–3 weeks, with Claude Code and agentic coding, Saito was able to build the MCP Server in a single day, making the 10x productivity goal a reality on this project. But there’s a key point to understand from Saito: 

“I could build the Treasure Data MCP server in one day because I know Treasure Data’s internals and APIs, and I have the expertise needed for understanding the generated code. We still need to have a lot of knowledge, both for Treasure Data and programming languages, but I was able to save a lot of time using Claude Code.”

Using Claude Code beyond the engineering team

Claude Code can support more than just the engineering team, and Treasure Data is exploring different use cases across various teams. For example, the support team utilizes Claude Code and the MCP Server to investigate customer activities, such as the issues customers are experiencing in their workload. 

Before Claude Code, the team had to write SQL queries or review customer usage logs. Today, the MCP can write the queries for them, enabling faster answers to questions. 

With a goal of improving productivity 10x, the team is now measuring how fast tasks are being completed now compared to before Claude Code. In GitHub, they can view the history of when a ticket was created and completed, allowing them to create a dashboard that calculates the average time spent on tasks. 

Another metric is to examine the amount of code written by Claude Code, as reflected in the commit history of the Treasure Data MCP Server.

The future of AI-driven development at Treasure Data

Earlier this year, only 20% of Treasure Data engineers had adopted agentic coding tools. Today, that number has increased to over 80% — and it’s ticking up by the day. 

Claude Code empowers engineers to either create code with minimal knowledge (“vibe coding”) or manage complex, end-to-end development workflows. By functioning as both an educational tool and a productivity booster, Claude Code stands out as a superior option compared to other AI tools.

As they move forward, Treasure Data’s engineering team looks forward to seeing these tools get faster and deliver higher-quality code. Engineers are still writing most of the code, but having Claude Code is like having a code assistant always beside them, helping along the way.