Tasks as Code: Move Your Tickets into the Repository

“Tasks as Code” proposes keeping development tasks as Markdown files in Git repositories rather than in separate ticketing systems. The case for this approach is especially strong with AI coding agents: when requirements, code, and documentation are accessible together, agents can inspect the task, understand the system, plan changes, and report what they did. A task file can be brief for routine maintenance or include acceptance criteria, dependencies, technical decisions, and tests for complex features. The suggested workflow is to describe the task, have the agent analyze it and draft a plan, let a developer review and approve that plan, then implement the change and add a report to the same file. This preserves requirements, decisions, and results in reviewable Git history while keeping people in control of implementation.

One example keeps code, documentation, and tasks in separate repositories, with the latter two cloned into the code directory but excluded from its Git tracking. Separate repositories allow different access rules, such as keeping sensitive or unreleased tasks private. Developers and agents can work in an IDE, contributors can use GitHub, and nontechnical colleagues can use a browser-based CMS; Git remains the system of record. Task histories can also help generate release notes and identify documentation updates. The approach requires only Markdown, Git, and an agent, though interfaces can make collaboration easier. The desired tools are Git-native, file-based, self-hosted, able to manage tasks and documentation together, and open to different AI providers. It also raises whether conventional ticket-system features remain necessary when agents handle more work.