Accelerating Innovation: The Use of AI in R&D at VMS Software
Within VMS Software's R&D organization, the use of AI has transitioned from experimentation to everyday practice, with engineers firmly integrating AI into their workflows. Getting to this point, however, was far from easy due to the fact that most off-the-shelf AI systems don't know OpenVMS and related technologies well enough to be reliably useful. In order to get to where we are today, we had to do significant groundwork to build a proprietary domain knowledge base as well as OpenVMS-specific AI plugins and skills. In this article, I would like to outline how our engineers use AI today and spotlight some of our most promising AI projects.
AI Usage in Everyday Tasks
At the time of writing this, the most widespread AI adoption use cases among our R&D teams are as follows:
Code comprehension. For engineers encountering an unfamiliar codebase or language, AI provides explanations and contextual guidance that can compress hours of learning into minutes. This proves especially valuable when working with legacy systems with sparse or outdated documentation.
Code refactoring, review, and vulnerability testing. Engineers use AI tools to assist with refactoring across large or unfamiliar codebases. AI is particularly effective at suggesting correct syntax, idiomatic patterns, and consistent naming. For new and changed code, we use AI tools and agents to perform code review, suggest improvements, and highlight possible vulnerabilities.
Test creation. Teams use AI to draft unit and integration tests, generate test templates, and translate existing tests between scripting languages (shell, Python, DCL). This noticeably reduces the time required to build out test coverage for new and legacy code.
Automation scripting. Engineers use AI to write automation scripts across multiple languages.
Drafting documentation. AI is used to draft internal documents, produce summaries of meetings and Jira tickets, and improve the clarity of written communication.
It is important to point out that all AI usage at VMS Software always adheres to a vital principle: any and all AI-generated output must be reviewed and verified by a human expert. AI may accelerate the work, but it does not replace human judgment.
Active AI Projects
I would now like to introduce some of our most ambitious and promising AI projects. As you will see in a moment, we don't use AI because it's fashionable – we use it to help us solve real-world problems that are relevant to us and our customers.
Project 1: AI-Powered Knowledge Base
The Problem
VMS Software has accumulated decades of domain knowledge: product documentation, resolved customer support cases, and internal engineering wikis. Historically, that knowledge was spread across several resources and cumbersome to search.
The Approach
This project consolidates our entire body of domain knowledge into a structured knowledge base purpose-built for LLM retrieval. The knowledge base is specifically designed to compensate for the lack of OpenVMS knowledge in off-the-shelf LLMs. When asked an OpenVMS-related question, AI retrieves relevant source material from the base and uses it as context, grounding responses in our own trusted knowledge.
As bugs and customer cases are resolved and new documentation is produced, the knowledge base is expanded, ensuring that AI's responses reflect the current, up-to-date state of knowledge rather than a historical snapshot.
Strategic Significance
This project is the most strategically significant one in our portfolio because it builds the infrastructure for almost all other AI projects.
Project 2: Autonomous Open Source Porting Assistant
The Problem
The OpenVMS ecosystem relies on up-to-date open-source software (libraries, tools, runtimes) that is maintained on Linux. This poses two problems. Firstly, porting each package to OpenVMS has traditionally been a manual effort and required deep expertise in both Linux and OpenVMS. Secondly, open-source projects update too fast for any porting team to keep up with. As a result, a backlog inevitably builds, forcing us to make difficult choices about which packages to prioritize.
The Approach
This project builds a specialist AI agent equipped with the domain knowledge (OpenVMS internal documentation, historical porting patterns, and common failures) needed to speed up porting work. The agent can compile, test, and validate its own output against a real OpenVMS environment.
Strategic Significance
The project steadily closes the package availability gap between Linux and OpenVMS and allows the open-source porting team to work at a pace that was not considered possible before.
Project 3: AI-Assisted Modernization of Legacy Codebases
The Problem
VMS Software maintains a portfolio of legacy products built on technology stacks that predate modern development practices. Since the codebases of those products are often large, sparsely documented, and only fully understood by a handful of senior engineers, modernizing those products (migrating them to newer languages, frameworks, or architectures) has historically been considered too expensive and risky to justify.
The Approach
Given an existing codebase, this AI system can construct a working model of its structure and intent, identify candidates for refactoring and migration, and suggest ways to modernize. The output is a series of structured guidelines that allow a human engineer to make informed decisions much faster than through manual code review.
AI can also produce baseline documentation for undocumented code, thus making the corresponding product accessible for a wider audience of new engineers.
Strategic Significance
The main bottleneck of legacy modernization projects is usually code comprehension, which is exactly what many modern LLMs excel at. This project protects the long-term value of our product portfolio and opens a path to modernization that would otherwise be economically impractical.
Project 4: AI Agent Plugins for OpenVMS Tooling
The Problem
AI coding assistants have proven their effectiveness in mainstream development environments such as Python, JavaScript, and common Linux toolchains. However, as stated above, OpenVMS and related technologies are often outside the AI comfort zone.
The Approach
This project builds plugins and skills that give AI agents the ability to directly operate the standard development tools used by VMS engineers. Initially aimed at integrating AI with VMS/XDE (our main development tool), the project's ultimate goal is a general capability layer – an agent that can compile OpenVMS applications, run their test suites, interpret output, and iterate. Having this capability at the tooling level makes it a building block for higher-order agents that can write code and validate it against a real OpenVMS execution environment as part of the same workflow.
Strategic Significance
Similar to our AI-powered knowledge base, the primary value of this project lies in the groundwork it lays, enabling broader AI initiatives that depend on an AI's ability to interact with OpenVMS tooling.
Project 5: System Dump Analysis
The Problem
System dump analysis on OpenVMS has always been a specialist skill requiring both tool knowledge and domain expertise. Unfortunately, many engineers who possess such expertise are now in retirement, and new hires can't become experts overnight.
The Approach
This project builds an AI system that can parse system dumps, identify relevant data, and produce a human-readable summary of the system state at the time of the event. While such a summary cannot replace human analysis, it can significantly reduce time to first insight and enable engineers without specialist dump analysis skills to triage issues independently.
Strategic Significance
Reducing the time and expertise required to interpret a system dump noticeably shortens incident response time and reduces the load on the small number of engineers with the relevant expertise. As this capability matures, each analyzed dump becomes a resolved case that contributes to the OpenVMS knowledge base and can inform future troubleshooting.
Conclusion
As I stated in the beginning, AI is now in daily use across our company's R&D organization, helping our engineers alleviate the burden of mundane tasks and focus on innovation. All of our AI projects are business-driven and address real-world problems and challenges that we and our customers are facing.
Speaking of challenges, there are still a number of them to overcome. Choosing the right AI tool for a given task, writing prompts that produce useful output, and leveraging AI agents are skills that take time to develop. We are investing in ongoing training for engineers to build AI, LLM, and tooling literacy, and we treat these skills as core engineering competencies rather than optional supplements.
The opportunity ahead is large. The foundation laid today translates into reliable development practices and operational advantages.
Chief R&D Officer
