An Honest Review of AI Programming

5 hours ago

AI adoption is often mandated by management without engineering background, causing resistance, but the author found it useful for search and summarization.
Hallucinations are inherent in LLMs; always verify primary sources, especially when using AI for coding or research.
LLMs are poor at writing code, especially for game development, due to limited and poor-quality training data, often producing over-engineered solutions.
AI coding assistants can help explore unfamiliar codebases but require careful double-checking to avoid wild goose chases.
The AI business model is currently unsustainable; costs may rise, and the value of AI tools should be weighed against traditional solutions.

AI assistants are most valuable for reviewing code and finding bugs, with frontier models performing well.
Refactoring tasks benefit from AI, reducing the cost of fixing design mistakes, but review can be challenging due to mixed-in incorrect changes.
Writing code together with AI is problematic because AI makes poor decisions and often disregards instructions, though future harnesses may improve collaboration.

Traditional tools like Vim and Emacs gain trust through deep customization and muscle memory, making them feel like extensions of the developer.
AI coding agents speed up output but reduce trust; usage rises while trust falls (from 40% to 29% in a recent survey).
New AI tools require a shift in process and culture, not just tool adoption; a broken process remains broken even with better tools.

<antirez>8 days ago

The author feels a responsibility to help less experienced programmers adapt to AI-driven changes in programming, based on their own foresight and experience.
The key shift is that programmers should focus on controlling the ideas and design of software rather than reading code line-by-line, because LLMs handle local optimization well and generate large amounts of code.
Reasons to prioritize ideas: LLMs produce verbose code, humans have limited time, and it's more efficient to evaluate design through prompts instead of scanning functions.

AI coding tools benefit experienced developers more than novices, creating a 'skilled orchestrator paradox' where expertise is needed to use tools but the tools may prevent expertise formation.
Studies (JetBrains, UPenn, Anthropic) show that heavy reliance on AI for code generation leads to 'illusion of competence,' skipped planning, and worse learning outcomes; mitigated use improves learning.
Novice developers who avoided AI assistance performed better, developing 'negative expertise' to ignore bad suggestions, while those who relied on AI often got lost and failed to build true understanding.

Invest in upfront alignment by making implicit assumptions explicit in the initial prompt to avoid unexpected AI outputs.
Restart with better context rather than steering AI when initial outputs are far off; avoid compounding errors through path dependence.
Equip AI agents with the same tools you use (CLI, MCP, browser control) to handle setup, testing, and instrumentation tasks.

AI assistants are most valuable for reviewing code and finding bugs, with frontier models performing well.
Refactoring tasks benefit from AI, reducing the cost of fixing design mistakes, but review can be challenging due to mixed-in incorrect changes.
Writing code together with AI is problematic because AI makes poor decisions and often disregards instructions, though future harnesses may improve collaboration.

Traditional tools like Vim and Emacs gain trust through deep customization and muscle memory, making them feel like extensions of the developer.
AI coding agents speed up output but reduce trust; usage rises while trust falls (from 40% to 29% in a recent survey).
New AI tools require a shift in process and culture, not just tool adoption; a broken process remains broken even with better tools.

<antirez>8 days ago

The author feels a responsibility to help less experienced programmers adapt to AI-driven changes in programming, based on their own foresight and experience.
The key shift is that programmers should focus on controlling the ideas and design of software rather than reading code line-by-line, because LLMs handle local optimization well and generate large amounts of code.
Reasons to prioritize ideas: LLMs produce verbose code, humans have limited time, and it's more efficient to evaluate design through prompts instead of scanning functions.

AI coding tools benefit experienced developers more than novices, creating a 'skilled orchestrator paradox' where expertise is needed to use tools but the tools may prevent expertise formation.
Studies (JetBrains, UPenn, Anthropic) show that heavy reliance on AI for code generation leads to 'illusion of competence,' skipped planning, and worse learning outcomes; mitigated use improves learning.
Novice developers who avoided AI assistance performed better, developing 'negative expertise' to ignore bad suggestions, while those who relied on AI often got lost and failed to build true understanding.

Invest in upfront alignment by making implicit assumptions explicit in the initial prompt to avoid unexpected AI outputs.
Restart with better context rather than steering AI when initial outputs are far off; avoid compounding errors through path dependence.
Equip AI agents with the same tools you use (CLI, MCP, browser control) to handle setup, testing, and instrumentation tasks.

Hasty Briefsbeta