Hot take: the agentic workflow is deeply wrong

An experienced developer, with 30 years of coding experience, argues that the agentic workflow is deeply flawed for established, mature codebases. While acknowledging its utility for new projects or unfamiliar languages, the developer states that agentic coding often produces spaghetti code that violates architectural principles and is expensive to maintain. They suggest that for large codebases, direct interaction with chatbots or tools like aider, where the developer guides the AI on specific files, is more effective than fully autonomous agents.

This developer's 30 years of experience and two years of testing every agentic coding harness offer a rare, long-term perspective on the limitations of agentic workflows compared to their benefits for new projects.