Generative AI as Junior Developer? A Tech Veteran Says 'Baffling Mischaracterization'

Daniel Terhoren North, host of Modern Software Engineering, has expressed bewilderment at the increasing comparison of generative AI agents to junior programmers, stating it’s “nothing like” his three decades of experience with entry-level developers. North emphasizes that junior programmers, often in a state of conscious incompetence, are characterized by a genuine eagerness to learn, curiosity, and the ability to ask insightful questions, retain knowledge, and apply feedback. They thrive on “breakable toys,” pair programming, and actionable feedback, driven by the inherent human desire to contribute and grow. This fundamentally contrasts with the operational model of current AI.

In stark opposition, North characterizes generative AI agents as “transactional mercenary contractors with a severe memory problem,” coining the term “Colin the Contractor.” While possessing encyclopedic knowledge, these agents often hallucinate details, operate on a “billable token” model for every interaction, and exhibit no inherent “care” for codebase, product, or users. Their performance can be inconsistent, and crucially, they suffer from an incurable memory loss (compaction), forgetting instructions even mid-task, leading to unpredictable and potentially destructive actions. John Willis Botchaloop’s observation that agents possess “unlimited knowledge, unlimited speed, and no accountability” underscores their inherent risk profile. Effective engagement with AI requires small, well-defined steps, frequent validation, and external checking, as they cannot reliably self-assess. North posits that the miscomparison may stem from a transactional view of junior developers or a desire to replace them, advocating instead for AI as an augmentation tool and urging professionals to “treat the machine like a machine and the people like a people.”