Software engineers are reporting a significant decline in their ability to understand and maintain codebases after a year of operating under mandates to use AI for nearly all development tasks. This shift from manual engineering to a cycle of prompting, waiting, and reviewing is creating a feedback loop where developers feel they are learning less while the underlying software logic becomes increasingly opaque.
The Shift to Passive Development
The original post on Reddit describes a landscape where, a year after adopting a '100% AI mandate,' teams find that nobody truly understands the code being produced. The poster suggests that this reliance on AI-generated output has led to a situation where the logic of the software is no longer transparent to the people responsible for it.
Commenters on Hacker News echoed these concerns, with one developer noting that a peer at a large firm has never dreaded their job until now. This peer reportedly spends 10 hours a day entering prompts into Claude, waiting for results, and reviewing output without learning anything new, leading to a decrease in motivation to become a better engineer.
Skill Decay and Individual Impact
A primary concern raised in the discussion is the erosion of individual engineering skills. One commenter noted that they spent two days on a task that would have taken an afternoon in 2024, attributing this to a loss of cognitive ability to learn from mistakes and build upon manual knowledge.
The consensus among several participants is that AI usage currently benefits the company's immediate output but may be detrimental to the individual's professional development. This creates a trade-off where the speed of work is gained at the expense of the engineer's ability to maintain deep expertise in the codebase.
Design vs. Code as Business Value
The discussion also touched on the distinction between code as a product and software design as the 'engine' of value. One commenter argued that non-technical management often prioritizes AI-generated code because it is visible, while the intangible work of proper software design is harder to justify as a business expense.
Another perspective raised was the long-term risk to startup value. If a product is entirely written by AI, the business's value may decrease because the underlying logic can be easily reverse-engineered, leaving only employee expertise and customer relationships as unique assets.
Economic and Technical Trade-offs
Some participants questioned the timing of these mandates, suggesting that 'reverse token value inflation' means that waiting for cheaper, more powerful models might offer a late-mover's advantage. This suggests that going 'all in' on current AI models might be less efficient than waiting for the technology to mature.
Technical concerns were also raised regarding the quality of AI-generated code. One commenter questioned how to handle codebases that 'run like shit' when they are entirely AI-written, noting that even human developers who struggle with code rarely produce the redundant, messy outputs sometimes seen in large-scale AI generations.