GitHub’s legal department is not full of engineers. It is made up of lawyers, program managers, and business professionals. Yet several of them have used GitHub Copilot CLI to build internal tools that automate the repetitive parts of their work — contract drafting, DMCA triage, risk assessment, and more. Their experience suggests the main barrier to building AI tools may not be technical skill, but the assumption that technical skill is required.

The approach is straightforward: describe what you want in plain language, point Copilot CLI at a repository, and iterate. Two first-person accounts from the team illustrate how that plays out in practice.

Drafting contracts with a personal style guide

Ngandu Kasuku, Principal Product Counsel, found himself handling a large volume of partnership deals involving data, infrastructure, and product integrations. The deals varied enough that each one felt like starting from scratch, making the drafting process slow and repetitive.

He had been using Copilot CLI to handle tasks individually, but realized he could build something more durable around his own workflow. The result was terms-ai, a contract drafting tool. The project is scaffolded around a repository containing key documents: drafting resources, instructions, and workflows that guide the AI. This structure made outputs more consistent and reduced the copy-pasting between prompts that had previously slowed him down.

A central piece is an internal drafting style guide based on plain-language principles. Kasuku, who has preferred clear language since his commercial law days, incorporated the approach of the plain-language legal drafting movement into the guide. He also added a library of previously completed agreements. Because the agreements are sensitive, they stay in an approved, access-controlled internal environment — only the tool and its general workflow are open source.

Kasuku reports that the tool has cut his review and drafting time roughly in half. Provisions are more consistent across agreements and drafts now match his preferred style. More importantly, he says the lesson was not that AI can draft faster, but that he could build a tool reflecting his own judgment and way of working.

Jesse Geraci, Online Safety Counsel, started with a focused problem: analyzing source code to evaluate DMCA notices. The initial project was a set of Copilot instructions for recurring tasks — DMCA triage, code comparison, license checks, and circumvention review. The goal was to turn ad hoc prompt work into repeatable processes that a legal team could rely on for consistent fact-gathering and analysis.

Geraci was surprised at how little traditional engineering was needed. The core of the workflow was plain-language files: instruction sets, policy reference materials, and report templates. His legal drafting skills translated directly into building structured decision-making into the workflow. Over time, the project grew to include different analysis modes — clients get faster outputs with escalation recommendations, while lawyers get deeper review and both-sides arguments — and integration with external data sources.

That foundation eventually became a full desktop application for running predefined legal workflows in a clean interface. Building the app required substantial code, but the workflow instructions themselves remain editable in plain language within the app. The tool has since expanded beyond DMCA analysis to cover contract review, NDA triage, risk assessment, compliance checks, and response drafting. Under the hood, work routes through reusable skills and agents such as intake, playbook alignment, risk scoring, evidence verification, escalation routing, and report assembly. The key design point, Geraci emphasizes, is not technical sophistication — it is that legal teams can control behavior through readable Markdown.

Geraci’s takeaway: you do not need to wait for vendor software or become a developer to build serious AI tooling. If you can define your methodology, standards, and output format, Copilot can operationalize that knowledge. He is clear that the tool is a structured decision-support system, not a replacement for human legal judgment — it is designed to keep human review central while making analysis more consistent and scalable.

What this means for non-technical teams

Both accounts share the same pattern. Pick one repetitive task that slows you down. Use Copilot CLI to describe the problem and the desired solution in natural language. Store the relevant documents and guidelines in a repo. Iterate from there.

The tools these legal professionals built are not side experiments. They are actively used by teams, they cut contract drafting time roughly in half, and they turned people who thought they “could never code” into people who “just built something.” The habit spreads on its own, according to the team.

Many jobs outside traditional software development include repetitive tasks that are candidates for this kind of automation. The skills needed to start are ones many knowledge workers already have: the ability to articulate a process precisely, define standards, and describe output formats.