AI coding agents tend to produce solutions that are more complex than necessary to get the latest feature working. They often add abstractions, layers, dependencies, generalized frameworks, and common design patterns that satisfy the immediate requirement while making the system harder for both humans and future agents to understand, modify, and maintain.

This talk explains why great engineers must fight that complexity before it enters the codebase. Agents tend to draw from common patterns in their training data, even when the codebase needs a simpler solution. They also aren’t consistent about which patterns they choose across iterations. One session may solve a problem 1 way, while the next solves a similar problem with a different abstraction, naming style, dependency structure, or decomposition strategy. Over time, those inconsistencies erode conceptual integrity.

The session shows how AI-generated complexity enters production code through duplicate concepts, unnecessary service decomposition, shallow modules, inconsistent patterns, and generalized solutions for problems the system doesn’t actually have. It also shows how teams can build a simplicity discipline into their workflow: refactor for simplicity before review, make complexity carry the burden of proof, design modules with small interfaces and substantial internal functionality, and organize code so engineers and agents can navigate safely.

Attendees will leave with a practical model for identifying overbuilt agent output, challenging new abstractions, collapsing unnecessary layers, removing unjustified dependencies, reconciling inconsistent patterns, preserving conceptual integrity, and requiring the simplest good solution before merge.

This talk directly references the author’s Engineering Standards for Agentic Software Development, especially RF4, RF6, RF7, RF9, SI1, SI2, and SI3: https://www.linkedin.com/pulse/engineering-standards-agentic-software-development-edensoft-park-ki1se

Target audience:
Senior engineers, staff engineers, architects, tech leads, and engineering managers focused on maintainability and software craftsmanship.

Preferred Session Duration:
50 mins including Q&A