The AI Native Dev - from Copilot today to AI Native Software Development tomorrow AI-First Project Management for Developers | Alex Gavrilescu on Backlog.md
16 snips
Oct 14, 2025 Alexandru Gavrilescu, lead engineer and creator of Backlog.md, dives into innovative AI-driven project management. He discusses the crucial role of clear specifications for effective AI coding and how persistent agents can autonomously manage tasks. The conversation highlights the need to re-evaluate Agile methodologies in light of AI's capabilities, emphasizing the shift towards disposable software and faster prototyping. Alex also explores the balance of human review and automated processes, forecasting a future where agents continually coordinate tasks and improve efficiency.
AI Snips
Chapters
Transcript
Episode notes
Teach Agents How To Use Your Tooling
- Provide agents with CLI instructions so they can create, edit, and manage tasks programmatically.
- Configure agent instruction files when initializing a project so tools know how to call Backlog.md.
Backlog Was Largely AI-Written
- Alex claims 99% of Backlog's code was written by AI under his guidance.
- He only hand-authored the agent instructions and orchestration pieces.
Agents Infer PR-Sized Task Granularity
- Agents can learn ideal task granularity from large-scale PR training data and split features into atomic, deliverable PR-sized tasks.
- Humans should review specs and the agent's implementation plans before execution to avoid scope creep.