747s and Coding Agents
Carl Kolon explores the unexpected parallel between a seasoned 747 pilot's career plateau and his own experience with advanced AI coding agents. While these agents dramatically boost productivity by handling entire features, Kolon argues they erode the deep learning and skill development inherent in traditional coding. He posits that intentional hands-on practice may be essential to maintain programming acumen in an AI-dominated future.
The Lowdown
Carl Kolon reflects on how the advent of sophisticated AI coding agents has fundamentally altered his work, drawing a poignant parallel to a conversation with a 747 pilot who expressed a sense of stagnation in his highly specialized role. Kolon, a programmer working at an AI lab, acknowledges the immense productivity gains but grapples with the potential loss of deep understanding and skill acquisition.
- The author recounts meeting a veteran 747 pilot who, despite his passion, felt his career offered no further personal improvement or learning opportunities.
- Initially, Kolon used LLMs to assist with search and debugging, but recent coding agents can now autonomously implement entire features, significantly increasing his output.
- This increased productivity comes at a cost: the author no longer builds skills or acquires deep system knowledge as quickly, mirroring the pilot's sentiment of 'no improvement'.
- He argues that reviewing AI-generated code is less instructive than creating it, and intervening in a partially incorrect AI solution can be more disorienting than starting from scratch.
- Kolon dismisses 'prompting' as the primary new skill, emphasizing that deep domain knowledge remains crucial for making sound design decisions, but AI makes acquiring this knowledge optional.
- He suggests that programmers may need to intentionally engage in 'educational coding' by hand, or compare their own solutions with AI's, to maintain foundational understanding.
Ultimately, while coding agents are an undeniable force for productivity that programmers should embrace, Kolon warns that an over-reliance on them could lead to a generation of developers who are efficient executors but lack the profound, hands-on understanding that once defined their craft.