Critical Thinking Habits for Coding with AI (for Duc ka) (Andrew Stellman)（Z-Library）

# Critical Thinking Habits for Coding with AI — Reading Guide ## 【One-Line Pitch】 A practical field guide for developers and team leads who want to use AI coding assistants productively without sacrificing code quality, design judgment, or long-term maintainability—built around the five-habit Sens-AI Framework. ## 【Book Arc】 - **Opening (~0%–11%)**: Introduces the core paradox—AI tools promise speed but often deliver debugging time, technical debt, and skill gaps. Sets up the Sens-AI Framework (Context, Research, Problem Framing, Refining, Critical Thinking) as the solution. - **Early (~11%–28%)**: Defines "vibe coding"—the rapid prompt-iterate loop that feels productive but can disengage developers from design thinking. Explains why this approach fails in production work and how it creates a "cognitive shortcut paradox" for newer developers. - **Middle (~28%–44%)**: Dissects specific AI failure modes: hallucination, overconfidence, and the less-known "rehash loop." Uses a concrete example (a simple class ballooning into a miniature framework with interfaces, mocks, and dependency injection) to show how technical debt compounds. - **Middle (~44%–61%)**: Explains the mechanics behind rehash loops—context windows, token limits, and how LLMs predict text rather than understand design intent. Argues that repeated bad answers usually mean insufficient context, not a broken model. - **Late (~61%–end)**: Moves into the Sens-AI habits themselves, with case studies, checklists, and coaching prompts for applying the framework in real team settings and training environments. ## 【Key Takeaways】 - **Vibe coding has a hidden cost** (Early): The prompt-run-error-paste loop feels productive but lets developers disengage from design decisions, leading to code they don't understand and can't maintain. - **The "cognitive shortcut paradox" hits early-career developers hardest** (Early): New developers who rely on AI skip building the critical thinking skills they need to evaluate AI output—creating a self-reinforcing skill gap. - **AI-accelerated technical debt shows up in recognizable patterns** (Middle): Tightly coupled modules, "God objects," and over-engineered abstractions appear quickly in AI-generated code because it's produced without intentional design decisions. - **The rehash loop is a distinct failure mode** (Middle): Unlike hallucination or overconfidence, rehash loops occur when the AI keeps returning variations of the same flawed solution—prompt tweaks make no real progress. - **Context windows explain why AI gets stuck** (Middle): LLMs predict text based on a limited context buffer; when that context is incomplete or poorly framed, suggestions drift and repeat. The fix is better context, not more prompting. - **AI is biased toward adding code, not revisiting decisions** (Middle): Left to its defaults, AI-assisted development grows codebases haphazardly rather than responsibly—requiring deliberate human oversight. - **"Trust but verify" is the core review principle** (Late): The framework emphasizes active review of AI-generated code as a teaching tool and quality gate, not just a safety check. ## 【Reading Tips】 - **Skim the opening chapters** (~0%–11%) if you're already convinced AI tools have drawbacks—the real value starts with the failure-mode analysis. - **Deep-read the rehash loop section** (~56%–61%): This is the most technically insightful part, explaining *why* AI gets stuck and what that means for your prompting strategy. - **Pay attention to the over-engineering example** (~44%): The simple class → miniature framework progression is the clearest illustration of how vibe coding balloons complexity. - **If you're a team lead or educator**, focus on the late sections with case studies and coaching prompts—they translate the framework into actionable teaching strategies. - **If you're an individual developer**, prioritize the habits themselves and the warning signals (rehash loops, overconfidence, technical debt patterns) over the teaching-specific content. ## 【Coverage Limits】 The excerpts focus heavily on problem diagnosis (vibe coding breakdowns, rehash loops, technical debt) and framework rationale. Detailed walkthroughs of each of the five Sens-AI habits, specific checklists, and most case studies appear in later sections not fully covered by the sample. ##