The most expensive AI coding failures often begin before the agent writes a single line of code. The engineer acts on an incomplete requirement, an unverified assumption, an unstated constraint, or a vague definition of done. The agent then fills the gaps with plausible defaults from its training data. The result can look impressive while solving the wrong problem or damaging important quality attributes.

This talk introduces requirements interrogation as a critical discipline for agentic software development. Before writing a technical specification, the engineer must surface what is unknown, interview stakeholders, clarify edge cases, define constraints, and understand what must not be sacrificed. After that, the engineer can use the agent to interrogate the requirement further by asking it to identify missing assumptions, unclear boundaries, and questions that should be answered before implementation begins.

The session is useful for product engineering organizations because it connects AI coding quality directly to product clarity. Teams that skip requirements interrogation will generate more code faster, with more rework. Teams that build the discipline will improve both engineering speed and product alignment.

Attendees will leave with a practical 2-stage model: stakeholder interrogation first, agent interrogation second. The goal is to surface missing assumptions, constraints, edge cases, and quality attribute priorities before generation begins.

This talk directly references the author’s Engineering Standards for Agentic Software Development, especially RI1, RI2, RI3, SP2, SP3, and DK1: https://www.linkedin.com/pulse/engineering-standards-agentic-software-development-edensoft-park-ki1se

Target audience:
Product engineering leaders, product managers, engineering managers, tech leads, and teams using AI to accelerate feature delivery.

Preferred Session Duration:
50 mins including Q&A