AI Generated Architecture Decision Records (ADR) (adolfi.dev)

A developer demonstrated an effortless workflow for generating Architecture Decision Records (ADRs) by using an AI agent (Claude Code) to scan a codebase and auto-create ADRs from a simple template. The setup uses template.md for individual record structure and overview.md as an index; the agent writes each ADR into /ADR/records and populates overview.md with a table of titles, links, status, and dates. The agent was instructed to prioritize high-impact architectural decisions and, crucially, to run continuously via an agents.md rule that prompts ADR creation whenever code changes affect overall architecture. This approach matters because ADRs are high-value but often neglected documentation. Automating detection and consistent formatting removes the friction that causes teams to skip them, preserves context while it’s fresh, and makes architectural intent discoverable for onboarding, audits, and future design debates. Technically, the pattern is lightweight and extensible: code/config scans to infer decisions, templated file creation, and an index for navigation—ready to be integrated into CI/CD or developer tooling. The author reports the agent uncovered many previously undocumented decisions, illustrating how AI can systematically capture tacit architecture knowledge (with the usual caveat that human review should validate correctness and nuance).