Engineering management after the cost of code collapsed
The advent of AI has dramatically collapsed the cost of producing code, forcing a critical re-evaluation of established engineering management practices. This insightful piece distinguishes between assumptions that remain valid (human coordination, trust) and those that are now obsolete (code generation effort). It offers a nuanced perspective on the future of development, emphasizing accountability and judgment over mere output in an AI-augmented world.
The Lowdown
Karim Jedda's article, "Engineering management after the cost of code collapsed," argues that the fundamental shift in the cost of code generation due to AI necessitates a complete overhaul of traditional engineering management rules and assumptions. He posits that while AI makes plausible code cheap, it doesn't automatically mean faster development or obsolete practices like code review and documentation.
- Assumption Audit: Managers must audit practices based on their underlying assumptions. Practices tied to human interaction, trust, and verifying correctness remain vital, while those focused on code generation efficiency are now outdated.
- Truth vs. Hype: Be wary of claims of dramatic speedups. Real gains are seen in boilerplate and greenfield work, not necessarily deep systems, and honest instrumentation is crucial to measure actual acceleration.
- Metric Misdirection: Old metrics like velocity, PR counts, and tickets closed are now misleading. With code volume cheap, focus must shift to business outcomes and system health, treating code volume as a cost.
- The Persistence of 'Right': While mechanical verification (tests, types) is faster, semantic verification (does code meet business needs?) still requires human judgment. The constraint moves from checking speed to specification quality and human accountability.
- Junior Pipeline Problem: AI absorbing foundational tasks historically used to train junior engineers creates an
The Gossip
AI's Analytical Acumen
Commenters discuss the potential for LLMs to go beyond code generation and actually analyze code quality. Specifically, one user wonders if AI can predict the price of a code change or discern good versus bad architectural patterns, prompting a challenge to test this capability.
Evolutionary Engineering Enablement
One perspective posits that the impact of LLMs on software development is part of a continuous evolution, much like past technological shifts. The commenter suggests that AI's unique 'unlock' allows engineers to tackle problems previously unapproachable, fitting into the natural progression of tools.
Critiquing Content Creation
A critical comment targets the style of the article itself, lamenting that the 'cost of writing collapsed' has led to an abundance of blog posts perceived as generic 'middle manager's TED talk' material. This hints at a broader concern about the quality and originality of online content in an era of easy generation.