Duolingo implemented GitHub Copilot to address challenges with developer efficiency and code consistency across their expanding codebase. The solution led to a 25% increase in developer speed for those new to specific repositories, and a 10% increase for experienced developers. The implementation of GitHub Copilot, along with Codespaces and custom API integrations, helped maintain consistent standards while accelerating development workflows and reducing context switching.
Duolingo, founded in 2011, has grown to become the world’s most popular language learning platform with over 500 million users. The company’s mission extends beyond language learning to building the best education platform in the world and making it universally accessible. To achieve this ambitious goal, Duolingo employs approximately 300 developers who work alongside language learning scientists, machine learning engineers, and AI experts. The company’s CTO and senior engineering leadership explicitly describe their philosophy as using “engineering as a force multiplier for expertise.”
This case study, published as a GitHub customer story, documents how Duolingo has integrated AI-assisted development tools, primarily GitHub Copilot, into their engineering workflows. While the source is promotional material from GitHub, it provides useful insights into how a large-scale consumer technology company has approached AI-assisted code development in production.
Duolingo faced several interconnected challenges that hindered developer efficiency and mobility:
The company had grown from three primary repositories to over 400 as they transitioned to a microservice architecture. However, each repository had developed its own culture and pull request processes, creating inconsistencies that made it difficult for developers to move between projects. This fragmentation was compounded by the use of various third-party tools like Gerrit and PullApprove for code review, which further contributed to workflow inconsistencies.
Additionally, developers were spending significant time on routine tasks such as setting up development environments, writing boilerplate code, searching through documentation, and navigating unfamiliar codebases. These distractions took focus away from solving complex business problems and slowed down the company’s ability to expand its content and deliver on its core educational mission.
Duolingo’s solution centered on adopting GitHub Copilot, described as an “AI-powered pair programmer that provides autocomplete-style suggestions to developers while they code.” The tool was deployed organization-wide as part of their existing GitHub Enterprise infrastructure.
GitHub Copilot offers developers two primary interaction modes: starting to write code and receiving autocomplete-style suggestions, or writing natural language comments that describe desired functionality. A key differentiator emphasized by Duolingo’s CTO Severin Hacker is the contextual awareness: “GitHub Copilot is unique in the sense that it looks at the context of the rest of your work and incorporates that context into its recommendations. Other tools don’t have that contextual awareness.”
This contextual understanding is particularly valuable in a large enterprise environment with sprawling codebases. Hacker specifically notes that “a tool like GitHub Copilot is so impactful at large companies because suddenly engineers can make impactful changes to other developers’ code with little previous exposure.” This suggests the LLM underlying Copilot is able to analyze the surrounding codebase to provide suggestions that are stylistically and functionally consistent with existing patterns.
One notable aspect mentioned is the simplicity of deployment. According to CTO Hacker, “GitHub Copilot works with all of our other code development tools, and enabling it across the entire organization is as simple as checking a box.” This low friction deployment is characteristic of SaaS-based LLM tools that integrate with existing development infrastructure, though it’s worth noting this claim comes from a promotional context.
The case study identifies several key use cases for GitHub Copilot at Duolingo:
Boilerplate Code Generation: Senior engineering manager Jonathan Burket emphasizes this as a primary use case: “Boilerplate code is where Copilot is very, very effective. You can practically tab complete the basic class or function using Copilot.” This aligns with common patterns in LLM-assisted development where repetitive, pattern-based code is well-suited for AI generation.
Reducing Context Switching: The tool helps developers “stay in the flow state and keep momentum instead of clawing through code libraries or documentation.” This speaks to the cognitive benefits of having relevant suggestions surface automatically rather than requiring manual documentation searches.
Cross-Codebase Contributions: The contextual awareness enables developers to make meaningful contributions to unfamiliar repositories more quickly, supporting the organization’s goal of internal mobility.
Alongside Copilot, Duolingo has adopted GitHub Codespaces, a cloud-based development environment. This was initially driven by practical issues—some developers had problems running Docker locally on Apple M1 machines—but the benefits extended to broader standardization and efficiency gains.
The combination of Codespaces and Copilot creates a unified development environment where AI assistance operates consistently across all developers. Principal software engineer Art Chaidarun notes that “setting up Duolingo’s largest repo takes just one minute” with Codespaces, compared to hours or days previously. This rapid environment provisioning reduces barriers to cross-team collaboration.
Duolingo has built extensive customizations using GitHub’s APIs to standardize workflows across repositories. One Slack integration for code review notifications reduced median code review turnaround time from three hours to one hour. These integrations work in concert with Copilot to create a cohesive developer experience.
The case study presents several quantitative outcomes:
It’s important to note that these metrics come from a promotional customer story, and the methodology for measuring developer speed improvements is not detailed. The distinction between familiar and unfamiliar developers (25% vs 10% improvement) does provide some nuance, suggesting the benefits are more pronounced when developers are working outside their usual domain.
While this case study presents a positive picture of AI-assisted development, several caveats should be considered:
The source is promotional content from GitHub, so it’s expected to highlight benefits while potentially underemphasizing challenges. The case study doesn’t address common concerns about LLM-generated code such as quality assurance, security vulnerabilities in AI-generated code, or the potential for developers to accept suggestions without fully understanding them.
The productivity metrics, while specific, lack methodological transparency. How was “developer speed” measured? What was the baseline period? Were there other changes occurring simultaneously that could affect these metrics?
The case study also conflates improvements from multiple tools—Copilot, Codespaces, and custom API integrations—making it difficult to attribute specific benefits to the LLM-powered components specifically.
That said, the quotes from engineering leadership suggest genuine adoption and satisfaction with the tools. The observation that Copilot is particularly effective for boilerplate code aligns with broader industry experience, and the emphasis on maintaining “flow state” reflects a real cognitive benefit of well-integrated AI assistance.
From an LLMOps perspective, this case study illustrates several patterns for deploying LLMs in enterprise development environments:
The case study represents a common pattern where enterprises adopt LLM tools through established vendor relationships rather than building custom solutions, trading customization for ease of deployment and maintenance.
This podcast discussion between Galileo and Crew AI leadership explores the challenges and solutions for deploying AI agents in production environments at enterprise scale. The conversation covers the technical complexities of multi-agent systems, the need for robust evaluation and observability frameworks, and the emergence of new LLMOps practices specifically designed for non-deterministic agent workflows. Key topics include authentication protocols, custom evaluation metrics, governance frameworks for regulated industries, and the democratization of agent development through no-code platforms.
Codeium's journey in building their AI-powered development tools showcases how investing early in enterprise-ready infrastructure, including containerization, security, and comprehensive deployment options, enabled them to scale from individual developers to large enterprise customers. Their "go slow to go fast" approach in building proprietary infrastructure for code completion, retrieval, and agent-based development culminated in Windsurf IDE, demonstrating how thoughtful early architectural decisions can create a more robust foundation for AI tools in production.
A comprehensive overview of how enterprises are implementing LLMOps platforms, drawing from DevOps principles and experiences. The case study explores the evolution from initial AI adoption to scaling across teams, emphasizing the importance of platform teams, enablement, and governance. It highlights the challenges of testing, model management, and developer experience while providing practical insights into building robust AI infrastructure that can support multiple teams within an organization.
