GenAI-based coding agents belong in the Cynefin 'complex' domain because the relationship between prompt and outcome cannot be fully predicted in advance. Unlike traditional developer tools (which are clear or complicated domains with stable, analyzable rules), LLM-based agents produce emergent behavior shaped by prompt wording, model configuration, retrieved context, and codebase interactions. This has concrete implications: engineering teams must prioritize fast feedback loops, automated testing, and human-in-the-loop checkpoints over static policies. Leadership must abandon top-down rollout mandates in favor of iterative, experimental adoption. Internal development platforms should act as learning amplifiers rather than policy factories enforcing a single 'one true way,' since effective prompting strategies and context engineering patterns continuously evolve as models and tooling change.
