Agentic coding delivers impressive local gains. Agents refactor faster, generate cleaner code, and close tickets that used to take days. At scale, however, teams discover a different problem: systems remain technically correct while business behavior quietly degrades.

Pricing and distribution are where this failure shows up first.

In this session, I share hands-on experience applying agentic coding to monetization logic in large B2B systems. Agents performed well at refactoring rules, reducing duplication, and improving test coverage. Yet subtle semantic drift emerged: temporary discounts hardened into entitlements, regional constraints leaked across markets, and experiments became policy. Nothing failed loudly. Tests passed. Revenue numbers looked plausible. The system had simply stopped behaving as intended.

The root cause was not agent quality. It was missing context.

This talk explores why pricing and distribution are uniquely vulnerable under agentic coding and what actually works to scale AI-powered development safely in these domains. We will look at how weak domain boundaries give agents too much freedom, why feedback loops fail to catch semantic errors, and how to design context that constrains agents without killing autonomy.

You will leave with practical patterns for context engineering in agentic workflows, including separating pricing intent from execution, defining agent-safe surfaces, and building feedback loops that detect meaning loss rather than just correctness. This is not a tools talk. It is about making agentic coding reliable in the places where it matters most.