Hitachi streamlines development work and embraces generative AI with GitHub Copilot

Gathering GitHub Copilot expertise to amplify system development productivity

Since its founding in 1910, Hitachi, Ltd. (“Hitachi”) has enriched society by developing peerless technologies and products. In recent years, Hitachi’s Social Innovation Business has combined the latest information technology, operational technology, and products to promote social sustainability and happiness. Additionally, creating new value by harnessing generative AI has become a significant endeavor for the company. In May 2023, it established the Generative AI Center and drew attention on Investor Day by announcing a target to lift system development productivity by 30 percent by 2027.

GitHub Copilot is a key part of these generative AI projects.

“Of the many generative AI use cases that we gather and share, streamlining system development is a major theme,” explains Gaku Saito, Director of Hitachi’s Lumada Solution Hub Business Promotion Center. Hitachi has chosen GitHub Copilot as a key tool to accomplish this goal.

“While we have our own development methods and tools, avoiding vendor lock-in is important for our customers,” says Saito. “One of the most reliable ways to do this is GitHub, a potential global standard.”

These factors pushed Hitachi to deploy GitHub’s services internally in 2021. The scope of application started with use cases for specific customers using enterprise servers and expanded to the cloud. These efforts eventually extended to the use of GitHub Copilot.

83 percent of users complete tasks faster, 30 percent greater in some instances

“Our internal evaluation of GitHub Copilot started in October 2023,” says Akihito Mizoe, a Director of the Lumada Solution Hub Business Promotion Center. Hitachi recruited around 200 internal participants to evaluate GitHub Copilot over three to four months. During this time, the users in some instances completed two surveys.

The surveys used the SPACE framework methodology recommended by GitHub, Victoria University, and Microsoft Research. This framework combines several metrics to evaluate developer productivity from multiple angles.

“The survey measured six performance metrics,” explains Mizoe. “Over 70 percent of users responded positively in five of these metrics and 83 percent completed tasks faster. Besides automatic code generation from comments and comments/test code generation from code, users benefited from the suggestions for typedef declarations and names of functions and variables.”

Currently, Hitachi uses GitHub Copilot mostly for coding and unit testing. Productivity gains average 10 to 20 percent in these areas, reaching 30 percent in some cases. The coding process of creating an initial framework can be time-consuming and mentally taxing. However, GitHub Copilot addresses this aspect, says Mizoe.

Coding tends to be a lonely one-person task, but the constant presence of GitHub Copilot eases the stress of developers.

Some respondents also felt that GitHub Copilot would be useful for learning a new programming language.

“Many Hitachi designers know all the details of business, but aren’t experts in newer development languages,” says Satoshi Igarashi, Director of the Generative AI Technology Development Department. With Hitachi recently migrating from COBOL to Java and using Python to leverage generative AI, GitHub Copilot is proving to be an excellent tool in the language-learning process. “Conversely, some developers who didn’t know COBOL learned it with GitHub Copilot,” adds Mizoe.

The number of GitHub Copilot users at Hitachi grew rapidly during the evaluation period and currently sits at about 2,000. Mizoe says his immediate goal is to increase this number to 5,000.

Integration with Hitachi's own development frameworks boosts code generation rates

In addition to evaluating and promoting GitHub Copilot, Hitachi has also integrated the solution with its existing development framework “Justware”.

According to Igarashi, “We mainly use GitHub Copilot for pair-programming and supporting unit tests, but combining it with Justware, which supports development control, makes development of large-scale systems even more efficient.” With Justware, developers can generate rule-based, well-controlled skeleton code from detailed designs. “However, we can’t generate complex business logic from rules, so we have to meticulously record details in design documents, from which developers manually code from scratch. Our challenge was to reduce workloads and enhance efficiency,” he says.

To accomplish the goal, developers use source code generated by Justware as the basis for improving the accuracy of code with GitHub Copilot. Generating business logic is a priority for Hitachi.

“By using GitHub Copilot on manually coded aspects, we can add code descriptions to detailed design documents and use them with Justware’s skeleton code to receive business logic suggestions. Combining GitHub Copilot with Justware has lifted the code generation rate from 78 percent to 99 percent according to our validation application.

Hitachi is also leveraging its GenAI System Development Framework, which launched in 2024. This framework covers everything from upstream processes such as requirement definition and basic design to integration and system testing. Coding prompts automatically generated from detailed design documents serve as the basis for generating code with generative AI. Hitachi considers these results as a “first draft” and intends to improve accuracy with GitHub Copilot.

“We typically use waterfall-style development for our many mission-critical projects, but we’re encountering more projects requiring rapid prototyping and agile development,” says Igarashi. “Based on the nature of a project, we’ll be able to expand the approaches of development with the support of GitHub Copilot or in combination with Hitachi frameworks.”

Knowledge sharing focused on retention

Community knowledge sharing is another notable aspect of Hitachi's use of GitHub Copilot.

“We run knowledge-sharing initiatives in many fields and launched generative AI-focused community activities with volunteers in 2023,” says Saito. In April 2024, Hitachi systemized these activities by launching the Generative AI Practitioner Community.

The structure of these communities is one outstanding aspect. 

“The purpose of our communities is to help knowledge owners and users collaborate, and there’s always a moderator between the two,” explains Saito. “Moderators review, approve, and disseminate knowledge so that it’s systemized and of high quality. We’ve systemized over 1,000 use cases within Hitachi into about 200 units of knowledge.”

Expanding the use of generative AI is not the singular focus: an awareness of retention is also essential.

“We’ve set two primary KPIs for retention,” says Saito. The first is the number of core active users. The community has around 15,000 participants currently. 2,500 are core active users, with Hitachi aiming to increase this number to 6,000.

The second KPI is yield. Hitachi quantitatively assesses the yield from internal activities on profitability, with the goal of enhancing those results.

The enterprise is considering more initiatives that will strengthen DevSecOps through greater development efficiency. This includes early vulnerability detection by combining GitHub Copilot with GitHub Advanced Security to incorporate security at the detailed design stage. These processes will enable the company to better apply the Cabinet Cyber Security Center’s “Security By Design” concept.

“More AI agents will be applicable to development work in the near future, and I believe we’ll reach the 'Agentic World' that Microsoft proposes,” says Igarashi. “We’re looking forward to working with Microsoft to apply this progress to knowledge retention and business.”

* Justware is a trademark or a registered trademark of Hitachi, Ltd. in Japan and other countries.
* This document includes services (or solutions) only provided in Japan.