Uploaded February 2026 | Updated September 2026, 2 weeks ago

---

Developing with AI agents like Claude, Codex, or Cursor can quickly turn into a troubleshooting nightmare. All the best practices you've mastered: DDD, SOLID, or Extreme Programming, suddenly feel obsolete when a non-deterministic AI model takes the wheel.

In this talk, I'll share my journey through this chaos and how it led me to create Backlog.md, a tool that brings structure back to AI-assisted workflows. It won't solve every problem, but by breaking work into smaller, manageable tasks, it makes results more deterministic and easier to review.

We'll see Backlog.md in action, implementing the Spec-Driven AI Development flow in minutes to build features efficiently.

By the end, you'll be able to:
Understand AI agents and how to structure tasks with Backlog.md
Apply the Spec-Driven AI Development flow for more predictable workflows
Use a repeatable loop to build features almost on autopilot

Backlog.md's code was written 99% by AI agents, proving the process delivers great results for developers.

P.S. No prior AI experience necessary. If you've ever typed `git pull`, you're ready!