Alex Gavrilescu presents Backlog.md, a free open-source terminal-based project management tool designed to improve AI agent workflows. The talk covers his journey from unstructured AI prompting (~50% success rate) to a structured spec-driven approach using markdown task files with defined 'why', acceptance criteria, implementation plans, and definition of done sections, achieving ~95% success rate. Backlog.md stores tasks as markdown files with front matter, supports a CLI and MCP interface for agents, offers a terminal UI (kanban board) and web interface, and uses git for syncing across branches. Key workflow: create a PRD, decompose into small tasks, have agents write implementation plans, execute one task per context window session, and review at three human checkpoints. The tool has ~4,600 GitHub stars and 99% of its own code was written by AI agents.
Sort: