With the advance of AI toward the end of 2022, Leidos wanted to stay ahead of the curve. AI's potential was incredibly promising: Could it help them continue honing their technical expertise even more? Support them in helping customers adopt and use the newest technology to their advantage? Ensure digital modernization for even their most high-security customers? Rob Linger, AI Software Architect at Leidos, expanded on how they vetted AI coding assistants and the impact Cody is having on their organization and their customers.

Before choosing a solution, they needed to tackle three key hurdles:

1. Security: What would the coding assistant do with their data? Where would their information go? How, if at all, could the AI coding assistant protect their data from unnecessary and potentially risky exposure? Given Leidos' clientele across government and defense industries, security was the absolute highest priority.
2. LLM lock-in: Many solutions confine you to whatever LLM they're using, even though generative AI is constantly evolving — and quickly. How could Leidos have more flexibility to iterate and keep up with the changing landscape?
3. Context: Legacy code and massively complex systems are virtually impossible for any engineer to fully "know." Therefore, context is one of the most needed capabilities and not many AI coding assistants are up to the task. What one gets out of a coding assistant is only as good as the prompt and whatever other context it has. Which solution could “get to know” Leidos the best?

Leidos was committed to getting the ideal coding assistant into the hands of all of their developers, making their jobs easier and improving their mission outcomes for customers.

Leidos began with an analysis of current alternatives. What coding assistant could meet Leidos' remarkably high standards and get ahead of their customers' needs—nearly all with information security requirements ranging from sensitive to the highest levels of classification.

They considered a broad list of solutions, from the biggest names on the market to smaller, emerging competitors. Some were easy to eliminate due to a lack of security and privacy.

Leidos saw huge potential in Cody's context, something they found lacking in alternative solutions. Per Linger, “we noticed very quickly that the context many AI coding assistants pulled from was very limited. For the most part, it was your open tab in the editor, and that was it. But when you're working on a software development project of any type that's even slightly more complex, you're going to run into instances where you need to import data from another project or repo.”

And while other solutions don't have nearly enough technical documentation, Cody has code repos Leidos can sift through. It's much more transparent and open-source, offering a window into how Sourcegraph communicates with the LLMs.

With Cody, Leidos isn't held to a specific LLM. “Generative AI is a fast-moving field, and the best model that's out there today may not be the best model tomorrow,” says Linger. “Something better could come out tomorrow. With a lot of solutions, you're locked into an LLM and putting a lot of faith in that model to keep up with the pace of change. Using Cody means we can avoid that LLM lock-in.”

Ultimately, Cody empowers Leidos to focus on the work that matters most, allowing it to better serve its customers. “We have the freedom to move at the pace our customers need. That's invaluable.”

Leidos is using Cody in a few specific ways:

1. For its own research and development – Leidos invests over $130M in research and development. Software is critical to delivering nearly all capabilities to their customers, so the less time spent reading, interpreting and getting common software code elements to work, the more time spent building the actual solutions and technology advancements their customers need.
2. To improve its products available to customers – in many cases, Leidos is hired to deliver solutions to the customer. The faster Leidos builds and advances those solutions, the greater their overall ability to be their customers' go-to provider and succeed as a business.
3. To amplify existing contracts already using Cody – Many Leidos customers aren't able to invest or bring in this kind of technology on their own; they don't have the budgets or the acquisition pathways to invest or scale like Leidos can. Leidos constantly looks to transition new technology solutions developed via their research and development or being utilized on other contracts in the most economical and impactful way for existing customers.

Scaling was also on the engineers' minds. By capturing the needs and data flows of these three objectives, Leidos could not only leverage Cody internally within their own engineering projects, but also externally for customers in different sectors and with varying levels of security.

One of their bigger conversations revolves around the modernization and migration of legacy code. For example, migrating from Oracle to PostgreSQL once took a full sprint, if not longer. Cody got them 80% to 90% of the way there within minutes.

Technology moves fast, and Leidos leans on Cody to teach its engineers the basics of both old and new technology that they're not familiar with but need to understand for projects they're working on. Thanks to Cody, they're saving time writing documentation for their code, generating boilerplate in seconds, writing unit tests with unparalleled ease which in turn improves code quality, and debugging significantly faster than before.

Leidos' senior developers, in particular, have felt the difference. Guiding junior developers used to take up about eight hours of the week, easily. With Cody, they've cut this down to two. In fact, Leidos lives by one important rule of thumb:

Linger highlights something important: “It's not just about time savings. It's about how you're able to spend your time.”

Cody gives Leidos engineers more space to focus on what matters — something felt by its customers. In fact, one engineer says: