A conference talk by Simon Martinelli describing his experience building a volunteer management system using AI coding tools, which led him to develop a spec-driven development methodology he calls the 'AI Unified Process.' The approach centers on two core artifacts — system use cases and entity models — as the source of truth, from which code and tests are generated. Key insights include: using system use cases (from 1987) over user stories for richer behavioral context, building an 'agent harness' with skills and MCP servers to guide AI, avoiding AI-generated project scaffolding and dependency management, keeping system prompts minimal to reduce hallucinations, and favoring self-contained system architecture over microservices for better AI context sizing. The talk also covers reverse engineering existing systems into specs for modernization projects, and argues that the real bottleneck with AI-assisted development is requirements engineering, not coding speed.