Tech • AI • Robotics • Game
It’s not just you
6/10TL;DR
Developers’ biggest frustrations with AI in software engineering center on bloated tests, broken focus, exaggerated claims, low-signal output, weak ownership, and the risk that junior engineers are pushed to ship faster without learning fundamentals.
KEY POINTS
AI often makes testing worse
One recurring complaint is that AI-generated code can flood projects with excessive end-to-end tests, sometimes changing or adding hundreds at a time. That creates slow and flaky CI pipelines, especially in simple applications that do not need massive test coverage. The underlying issue is not testing itself but poor judgment about what should be tested and at which level.
Concurrency gains can become context-switching losses
AI can generate code faster than a person can write it, but it also introduces waiting time while tasks run, responses arrive, or edits are applied. That often tempts developers to launch several features at once, only to become disoriented across three or four concurrent threads of work. For many engineers, AI improves throughput only up to a point; beyond that, it degrades concentration.
Hype and fear of missing out distort decision-making
The pressure to adopt every new AI agent, framework, or workflow has become a source of frustration in itself. Tooling changes so quickly that hard claims about one “correct” way to work are often obsolete within weeks. The broader lesson is that engineers are unlikely to fall behind simply because they are not using the latest named workflow immediately.
Claims about AI replacing deep engineering remain overstated
A major source of tension is the gap between what AI can do today and the claims made for it. Current models can help inexperienced developers produce working software in familiar categories, but that is far from designing and operating resilient services at the scale of Cloudflare or Vercel. The difference between shipping code and building durable systems remains substantial.
Too much output creates its own productivity tax
Another frequent complaint is the sheer volume of low-value text generated by AI tools. Even when asked for a short answer or a list, many systems return verbose explanations, framing, and repetition that bury the useful information. That turns review into a time sink, with developers reading hundreds of words to extract a single actionable sentence.
Ownership improves when AI is used for exploration, not just delivery
A practical way to regain ownership is to treat AI-generated work as a prototype rather than the finished system. One effective pattern is to let AI help sketch a first version, learn from its structure and failures, then rebuild a second version deliberately with clearer architecture, fewer tests, and less code. In one example, a project moved from about 100,000 lines of code toward roughly 25,000 to 30,000 after that redesign process.
Junior engineers face the hardest trade-off
The most serious long-term concern is the effect on less experienced developers. Teams can now demand fast output from juniors because AI tools make them look immediately productive, but that speed can bypass the slow process of learning what good engineering looks like. Without time to study architecture, dependency management, edge cases, testing strategy, and trade-offs, newer developers risk becoming prompt operators instead of engineers.
Fundamentals still matter in hiring and career growth
Even if AI capabilities continue to improve, companies still need people who can reason about systems, challenge weak ideas, and make sound technical decisions. The stronger candidate is not the person who simply relays instructions into a model, but the one who can break down a problem, choose the right tests, shape the interfaces, and explain the trade-offs. AI may raise the floor for output, but judgment remains the differentiator.
CONCLUSION
AI is accelerating software work, but it is also exposing weak habits and creating new failure modes. The central divide is no longer whether to use AI, but whether it is being used to deepen engineering judgment or to replace it.
Ask a question