The most important AI coding work happens before generation begins. Great engineers don’t start by asking an agent to write code. They first understand the requirement, interrogate assumptions, surface constraints, define invariants, identify edge cases, and decide how the output will be independently verified. That preparation determines whether the agent produces real engineering leverage or fast, plausible waste.

This talk gives teams a practical preparation model for agentic development. It begins with stakeholder interrogation: clarifying what is actually needed, what is uncertain, what must never happen, and what quality attributes matter. It then moves to agent interrogation: using the agent to ask questions before implementation begins, exposing hidden assumptions and gaps in the engineer’s understanding. Finally, it covers specification discipline: translating the requirement into a clear technical plan before any code is generated.

The session is especially valuable for teams already experimenting with Cursor, Claude Code, GitHub Copilot, Codex, and similar tools but seeing inconsistent results. It shows how disciplined preparation reduces rework, improves verification, lowers review burden, and makes agentic coding safer for production systems.

This talk directly references the author's Engineering Standards for Agentic Software Development, especially RI1, RI2, RI3, TD1, TD2, TD3, SP1, SP2, and SP3: https://www.linkedin.com/pulse/engineering-standards-agentic-software-development-edensoft-park-ki1se

Target audience:
Engineering managers, tech leads, senior engineers, product engineering leaders, and product managers working closely with AI-enabled engineering teams.

Preferred Session Duration:
50 mins including Q&A