GitHub legal team built AI tools without coding

GitHub's legal and business teams—a group of lawyers, program managers, and non-technical professionals—faced mounting pressure from repetitive work. Ngandu Kasuku, Principal Product Counsel, found himself buried in partnership deals involving data, infrastructure, and product integrations starting around March or April. Each deal was different enough that every new matter felt like starting from scratch. After experimenting with GitHub Copilot CLI on individual tasks and seeing what others had built, Kasuku realized he could design a tool around his entire way of working.

He created terms-ai, a contract drafting application that uses Copilot CLI at its core. The tool stores key documents, instructions, and drafting resources in a repository, giving him a single place to organize and version everything that guides the AI. One of its main features is an internal drafting style guide rooted in plain-language legal principles—Kasuku had never understood why contracts needed words like "heretofore" and "therewith." The tool also maintains a library of agreements he had already completed, so when an existing partner sends a new addendum, it can draw on that earlier work. These agreements stay in an approved, access-controlled internal environment. The tool's workflow is open source, but sensitive agreements are not. Since launching terms-ai, Kasuku cut his review and drafting time roughly in half, improved consistency across agreements, and shifted to the plain style he preferred.

Jesse Geraci, Online Safety Counsel, started with a narrower problem: analyzing source code quickly to evaluate DMCA (Digital Millennium Copyright Act) notices. He began with a set of GitHub Copilot instructions for recurring tasks—DMCA triage, code comparison, license checks, circumvention review—and turned messy, one-off prompt work into something repeatable that the legal team could trust. The core "programming" consisted of plain-language files with workflow instruction sets, policy reference materials, and report templates. Geraci used his language crafting skills as a lawyer to build structured legal judgment into the workflow itself, rather than writing source code. The system grew to include different analysis modes for clients (faster outputs and escalation recommendations) and lawyers (deeper review and both-sides arguments), plus integration with external data sources. When he handed it off to the team, they immediately started using it and asked Copilot to do more.

That foundation evolved into a full desktop application for running predefined legal workflows. Building the app required writing code, but the core instructions that customize workflows remain editable in plain Markdown. The app now covers many in-house workflows: contract review, NDA triage, risk assessment, compliance checks, and response drafting. Under the hood, it routes work through reusable skills and agents (intake, playbook alignment, risk scoring, evidence verification, escalation routing, report assembly), but the crucial point is that legal teams can still control behavior in readable Markdown without becoming full-time developers. Geraci's key lesson was that he did not need to wait for a perfect software vendor or become a developer himself to build serious AI tooling—if you can clearly define your methodology, standards, and output format, GitHub Copilot makes it easy to operationalize that knowledge. Both authors emphasized that these tools are structured decision-support systems designed to keep human review central, making legal analysis more consistent, transparent, and scalable without replacing human judgment.