AI Coding assistants provide little value because a programmer's job is to think

Perceived Value and Common Use Cases

• Many commenters say AI coding assistants are already saving them “tons of time.”
• Effective uses cited:
  • Generating boilerplate, scaffolding, and repetitive variants of similar code.
  • Translating natural-language requirements into initial implementations.
  • Writing or expanding unit tests, including edge cases the developer hadn’t thought of.
  • Explaining unfamiliar code, functions, or modules; serving as a “rubber duck.”
  • Searching large codebases with natural language rather than manual grep/search.
  • Quick one-off scripts, data imports, CLI snippets, config migrations, etc.

Negative Experiences and Limitations

• Others report mixed or strongly negative results:
  • Frequent hallucinated APIs and nonsensical code, especially for niche languages (Zig), embedded/RTOS, complex C++/Qt, Rust, ElasticSearch, esoteric front-ends, and 3D/graphics.
  • Correcting hallucinations and mis-assumptions often takes longer than writing code directly.
  • Tools stumble on “long-distance” dependencies, intricate bugs, and complex domain logic.
• Some find AI more confusing than helpful, abandoning it after repeated failures.

Workflow, Context, and “Using It Right”

• Strong emphasis that results depend heavily on:
  • Model choice and recency.
  • Providing rich context (docs, entire files, repo-wide indexing).
  • Decomposing tasks into small, well-specified steps.
  • Iterating on plans and tests, not just asking for “make it better.”
• Supporters argue skeptics often use outdated models, give poor prompts, or expect end‑to‑end autonomy rather than “assistive” use.

Thinking vs Typing; Abstraction vs Boilerplate

• One side: typing isn’t the hard part; the valuable work is design, abstraction, and understanding – AI that just shovels more mediocre code “faster” adds little.
• The other side: by offloading rote coding and refactors, AI increases time available for real thinking.
• Several note that AI is good at patterns and boilerplate, much weaker at novel architectures and deep abstraction, especially in large, evolving (“Day‑50”) systems.

Skills, Jobs, and Future Trajectory

• Concerns raised about skill atrophy, loss of judgment, and commoditization of “thinking.”
• Others argue this is similar to past shifts (compilers, higher-level languages): most devs will eventually trust AI the way they trust compilers.
• Some predict employers will soon mandate AI use for productivity; others doubt that, citing brittleness and correctness concerns.

Overall View of the Article

• Many commenters feel the article’s claim that AI assistants provide “little value” ignores widespread, concrete productivity gains.
• Critiques focus on the author’s limited examples and apparent lack of real-world experience with modern tools, while acknowledging valid worries around hype, reliability, and long‑term impact on the craft.