Cantrill: LLMs Lack the Laziness That Forces Good Abstractions

Bryan Cantrill argues that LLMs are missing a critical engineering virtue: laziness. Because generating code costs an LLM nothing and it has no stake in future maintenance burden, it will keep piling complexity onto a system rather than collapsing it into cleaner abstractions. The result is software that grows larger and noisier, optimizing for surface-level vanity metrics while degrading on the qualities that actually matter.

The deeper point is that human laziness — really, the constraint of finite time — is what historically pressured engineers to design crisp abstractions in the first place. Nobody wants to keep paying the tax of a clunky interface they’ll have to use again tomorrow. LLMs don’t share that incentive, which means using them without discipline tends to externalize cost onto the humans who eventually have to read, debug, and extend the output.