The change passed every test. Here's what the AI coding agent couldn't see.

Context engineering for AI coding agents involves supplying not only repository-specific instructions and code but also the broader dependency context needed to assess changes across complex enterprise systems. A banking example illustrates how replacing single-letter account-status codes in a Java service could silently disrupt a COBOL fee batch, a PySpark lakehouse pipeline, and a finance dashboard that depend on the same Db2 column but exist in separate repositories and use different naming conventions. The text argues that repository search and language servers cannot reliably reveal these cross-language, configuration-driven, or runtime-generated dependencies, while a code-and-data graph can map relationships among tables, columns, jobs, services, datasets, dashboards, and owners. In tests using IBM’s CICS Banking Sample Application, the author reports that agents supplied with Foundational Context of Code and Data over MCP produced more useful dependency answers than those relying on local repositories or language servers alone. The proposed workflow combines rules files that tell agents when to check for impact, graph-based queries that identify downstream consumers before edits, and pull-request impact analysis to alert reviewers to affected systems, with the goal of reducing—not eliminating—the risk of AI-assisted changes breaking production dependencies.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.