Linways Case Study

The engineering team was carrying the routine load that slows every software organisation: reading unfamiliar code before it can be changed, writing the same shape of unit test repeatedly, and starting troubleshooting from a stack trace with no obvious entry point. The organisation wanted AI assistance to reduce that load, but only if it lived inside the workflow developers already used. A separate AI tool sitting outside the repository, the branch and the pull request would have been opened once and then forgotten. It also needed more than one model available, so developers could match model to task, and it needed real consumption visibility rather than an assumption that licences were being used.

GitHub Enterprise was established first as the development foundation, giving the organisation one consistent platform for repositories, branches, pull requests and code review, with access and governance configured before any AI was introduced. GitHub Copilot was then enabled as the AI layer inside that same workflow, so assistance appeared where developers were already working rather than in a separate tool. Anthropic Claude models became the primary assistance layer across code generation, code comprehension, debugging, refactoring, test authoring and technical documentation, with GPT and Gemini variants available alongside them. Every AI suggestion is reviewed and validated by a developer before it enters the codebase and travels through the standard pull request process. Model level AI credit reporting gives the organisation visibility into which capabilities are being used, by how many developers, and how that mix shifts over time.

Tech Stack & Integrations

• •GitHub Enterprise — repository, branch and pull-request foundation
• •GitHub Copilot — AI assistance inside the developer workflow
• •Anthropic Claude (Opus, Sonnet, Haiku) — code generation, comprehension, debugging, test authoring
• •Multi-model access — GPT and Gemini variants available alongside Claude
• •Access & Governance Controls — enterprise permissions, branch policy, review gates
• •Consumption Analytics — model-level AI credit reporting and adoption tracking

• •94% developer adoption (16 of 17) within the first reporting period
• •79.5% of AI consumption on Anthropic Claude models
• •Model-level visibility into AI usage across the engineering team
• •Multi-model environment with developer-led validation on every change
• •Enterprise GitHub foundation for repository governance and collaboration

Code comprehension time decreased by 55%, from 90 minutes to 40 minutes per unfamiliar module. First-pass code acceptance increased from 70% to 88% after AI-assisted development was introduced.