AI Coding: Harder Engineering

The emergence of artificial intelligence in software development has sparked a debate on the true nature of engineering tasks. While AI promises to accelerate coding processes, it simultaneously raises the bar for software engineering complexity. This evolution challenges developers to adopt a new level of discipline and expertise, transforming their roles and expectations.

Simon Willison, an influential figure in the developer community, highlights this shift, noting that while AI coding assistants speed up development, they paradoxically make the engineering process more demanding. Willison stresses that to harness the full potential of these tools, developers need "extraordinary discipline and knowledge," emphasizing that AI does not diminish the requirement for human expertise but rather redefines it.

The assumption that faster coding leads to easier work is gaining acceptance among developers. Hillel, a reader responding to Willison, succinctly states, "It doesn't get easier, you just get faster," emphasizing the crucial distinction between speed and ease. With AI coding tools, software engineers face increased scope and complexity, introducing new layers of decision-making, verification, and oversight.

Even with productivity gains, professionals often find themselves dedicating more time to project management, which can be attributed to the capacity to take on more ambitious projects rather than a reduction in effort. The enthusiasm surrounding AI's capabilities often leads to increased project scopes, as developers are willing to take on more challenging work.

AI coding agents demand skilled software engineers, amplifying the consequences of a lack of discipline rather than eliminating it. While AI handles coding, engineers must grapple with intricate decisions related to compatibility, potential disruptions, and preserving existing functionalities. The gains from AI come with increased responsibilities, where the time saved is often reinvested into addressing more complex problems.

The effectiveness of AI in reducing correction and verification work can enable previously impractical projects. However, organizations face the challenge of managing the expanded workload resulting from productivity gains. New research from UC Berkeley suggests that productivity gains from AI can lead to unsustainable workloads if not properly managed.

Employees in a 200-person technology company expanded their responsibilities, integrated AI prompting into idle moments, and managed more tasks concurrently. This shift, driven by excitement over new capabilities, led to a normalization of extra effort as standard performance. The critical issue lies in how organizations define success metrics in an AI-augmented environment.

While AI can facilitate the discovery of valuable work, strategic allocation of productivity gains is crucial. Organizations must ensure that the "gains" from AI do not merely disguise an escalating workload, maintaining a balance between increased ambition and manageable workloads.