One Line Splits Tasks in Two
The most useful dividing line for deciding whether to hand a review task to AI is: is it mechanically decidable, or does it need contextual judgment?
Mechanically decidable = there's a clear rule and you can conclude without context; needs contextual judgment = you must understand the business goal and weigh trade-offs to make a call.
Work Suited to Handing the AI
Any check that's rule-clear, repetitive, and doesn't require understanding the business is fast and stable to hand to AI (or automated tools): formatting and layout, obvious typos, whether a null check is missing, whether doc comments are absent, common security anti-patterns.
适合 AI 的：
- 这里的变量声明了却没用
- 这个函数缺返回值类型标注
- 这个除法没防 0
- 这段拼 SQL 有注入风险Work You Must Shoulder Yourself
Any question that asks "is this design good for the business" can only take AI's answer as reference: whether to split into microservices, whether this abstraction is premature, whether this change breaks a downstream team's dependency, whether this product logic matches the real requirement.
These all require the context in your head: business goals, historical baggage, team understandings. AI can't see these—it can only guess from the text you give it.
自测 · 学完检查一下
想真正动手做题、记进度、攒连胜？到互动课里练。
When dividing up review work, what's the core dividing line this lesson uses to classify?
答案：Mechanically decidable vs. needs contextual judgment
The core dividing line is whether a task "has a clear rule and is mechanically decidable" or "needs weighing in context."
Which task is best suited to hand to AI for automatic review?
答案：Checking for variables that are declared but never used
Unused variables are a rule-clear, mechanically decidable check—just right for AI; the other three all need business and architectural context.
Which task should you decide yourself rather than outsource wholesale to AI?
答案：Whether this architectural change breaks downstream teams' dependencies
Downstream impact must be judged with team-collaboration and business context the AI lacks; the other three are mechanically decidable checks suited to AI.
Judge: "whether to split into microservices" is hard to hand to AI mainly because it needs business goals, historical baggage, and other context the AI can't see.
答案：Correct
Architectural trade-offs depend on lots of off-scene context; the AI can only guess from the text you give, so such judgments rest with humans.
Judge: a check like "does this SQL string have an obvious injection anti-pattern," with rules to follow, is mechanically decidable and suited to hand to AI.
答案：Correct
Common security anti-patterns have clear rules, are mechanically decidable, and are exactly the batch-stable checks AI is good at.
Judgments like business intent and architectural trade-offs are hard to hand to AI fundamentally because the AI lacks the ____ you hold in your head (a term for the background of a situation).
答案：context
The AI can't see context like business goals, historical baggage, and team understandings—it can only guess from the text you give, so such judgments must rest with humans.