One observation is that AI lacks 'laziness,' one of the virtues of programmers.



In his blog, software engineer Brian Cantril discusses the danger of losing the virtue of 'laziness' that human programmers possess in software development in the age of AI.

The peril of laziness lost | The Observation Deck
https://bcantrill.dtrace.org/2026/04/12/the-peril-of-laziness-lost/

Larry Wall, known as the creator of the programming language Perl , lists 'laziness,' 'impatience,' and 'arrogance' as the three virtues of a programmer in his book ' Programming Perl .' Here, 'laziness' doesn't mean simply cutting corners, but rather 'the attitude of thoroughly considering better abstractions and simpler designs in order to reduce the work of oneself and others in the future.'

According to Cantril, this kind of laziness actually involves a great deal of intellectual effort. He considers time spent repeatedly thinking about problems in one's head without writing code—a kind of 'hammock-driven development'—to be necessary work to reduce future burdens.



On the other hand, Cantril states that over the past 20 years or so, the range of people who create software has broadened, and people who don't call themselves programmers have become involved in development. He argues that the irony and design philosophy embedded in the word 'laziness' have become less clear, and there is a growing trend to view 'writing a lot of code' as a measure of success.

The emergence of LLMs further accelerated this trend. Cantril points out that LLMs act like a 'muscle booster' for development cultures that boast about the amount of code, as they allow users to execute their development approach with greater force. He adds that even the entire DTrace is only about 60,000 lines depending on how you count it, emphasizing the dangers of simply accumulating lines of code. He says that it's like judging a work of literature by its weight, and that even a beginner should be able to see that treating the amount of code as an indicator of productivity is a mistake.

For example, Y Combinator CEO Gary Tan posted on X (formerly Twitter) that he wrote '37,000 lines of code a day.'



However, Polish software engineer gregorein investigated Tan's 'newsletter-blog-thingy' and found that a single load contained multiple test harnesses, a Hello World Rails app, a hidden text editor, and eight different variations of the same logo.



While Cantril acknowledges that individual problems are fixable, he sees the root of the issue not as the mistakes themselves, but as the fact that 'the methodology of generating large amounts of code using LLM tends to lead to the accumulation of unnecessary elements.' In other words, Cantril argues that LLM inherently lacks the programmer's virtue of 'laziness.' For LLM, the amount of work doesn't have the same cost as human effort, and there's no motivation to save time for oneself or others in the future.

Therefore, Cantril warns that leaving LLMs unchecked could not improve the system, but rather pile even more code on top of the existing mess. While it's easy to appeal to superficial metrics like lines of code, it risks undermining what's truly important, such as simplicity of design, low cognitive load, and future maintainability.

Cantril stated, 'It is precisely because humans have constraints such as time and cognitive load that we are motivated to seek clearer abstractions and simpler structures when faced with complex systems.' He added, 'Excellent engineering is born from constraints, and we cannot expect LLMs, who do not have the constraints of time and load, to spontaneously make the same judgments.'



However, Cantril is not rejecting LLM itself. He states that LLM is a powerful tool in software engineering and can be used to address technical debt and increase development rigor. Cantril emphasizes that LLM should not be treated as a substitute for human 'laziness,' arguing that 'LLM should be a tool that encourages good human laziness, and should be used not simply to increase code, but to create simpler, more powerful systems that will also benefit future developers.'

Related Posts:

in AI,   Software, Posted by log1i_yk