Do the Hard Work First, Then Use AI—Not the Other Way Around

The author describes a personal method born from engineering: solve a genuinely hard problem by hand the first time, using research tools but doing every step yourself. Once it works, write down exactly how you got there—including the wrong turns—while it is fresh in your mind. The second time the same problem appears, do not repeat all the hand work; instead, take those notes and automate the boring, repeatable parts into a small script, freeing your attention for decisions that actually need judgment. Only the third time, or when you know you will face the problem forever, build something that handles it fully automatically. By then you understand the problem well enough to trust a machine with it and, crucially, to notice when the machine gets it wrong.

For years the author thought this method was a way to go faster. It is not. It is a way to ensure that when you finally hand a task to a machine, you have already earned the understanding to catch it when it fails. That turns out to be exactly what AI makes easy to skip. The author frames this not as an argument against AI—the whole site exists to help people use these tools—but as a method for preserving the single skill most worth keeping: the ability to think, even when a machine offers to do it for them.

The research backing this argument centers on how learning actually works. Decades of studies show that people remember answers they generate themselves far better than answers they are simply handed; that retrieving information from your own head is a much stronger teacher than re-reading; and that learning that feels harder now tends to last far longer. The author carefully notes that these studies measured memory and learning, not critical thinking directly, and that no study has tested the exact four-rung method. But the research is clear: understanding you build yourself sticks better than understanding you are handed.

Critical thinking depends on this foundation. You can spot surface-level problems—asking who is behind a claim, what they gain, whether other sources agree—without doing the work yourself. But the deepest check, knowing an answer is wrong because you understand the thing itself, only comes from having done the work. Skip the doing, and you are left with just surface checks. This is why rungs one and two (do it by hand and write it down) come first: that is where the understanding is built.

When you skip straight to the answer, research points to predictable failures. People remember where to find things instead of the things themselves—sometimes called the "Google effect." More worryingly, when a machine usually gets it right, people stop watching closely and miss the errors it does make. A skill you stop practicing fades. A 2025 study by researchers at Microsoft and Carnegie Mellon found that the more people trusted an AI tool, the less critical thinking they reported doing; they shifted from doing the work to checking the machine's output, and sometimes skipped even that. A separate survey of 666 people found the same link between frequent AI use, cognitive offloading, and lower critical-thinking scores. The author treats these as strong hints rather than proof, since both are correlational and rest on self-reporting, and refuses to overclaim.

The author also notes the MIT "cognitive debt" study involving EEG sensors and ChatGPT-written essays, raising it only because it has circulated widely in headlines. It is a small, not-yet-peer-reviewed preprint that has already drawn formal methodological critique. The author uses it as an example of the exact habit the post is about: do not bank a confident claim just because it flatters what you already suspect.

The difference between this engineering ladder and simply using AI to skip the work is one rule: you earn the automation, and you never fully take your hands off the wheel. The analogy is calculators. Children learn multiplication by hand first, then use calculators afterward—and a kid who understands multiplication can use a calculator all day and still notice when the screen says something absurd. A kid who never learned cannot. But there is a difference of scale: a calculator does one narrow thing you can spot-check instantly; a general-purpose AI drafts the essay, builds the argument, and decides what matters—the whole chain of reasoning in fluent, confident prose much harder to eyeball for error. The wider the task you hand off, the more you needed to have done it yourself to judge the result.

For parents and educators, the author offers a practical ladder. First, let a child do the work by hand and let it be a little hard; the difficulty is the engine of learning, not the enemy. Second, make them show their thinking out loud—not "what is the answer" but "how did you get there and where did you get stuck?" Third, introduce AI as the second pass, not the first draft. Flip the usual order: do the work first, then bring in the AI to critique it, poke holes in it, or offer a version to compare against. "Here's my answer; what did I miss?" builds judgment. "Give me the answer" replaces it. Fourth, keep one question permanently on the table: did the machine get it right? The method scales with age—a six-year-old's "doing it by hand" is sounding out a word; a sixteen-year-old's is sketching the argument before writing it. Match the ambition to the child. None of it requires technical skill.

The essay sits at the intersection of cognitive science and practical tool use. The author cites decades of learning research—the generation effect, retrieval practice, and desirable difficulties—to argue that hands-on struggle is not a cost paid on the way to understanding; it is the mechanism by which understanding forms. This is why the order of the four-rung ladder matters: automation works safely only after you have earned the knowledge to judge whether it is right.

The essay acknowledges both the power of AI and the genuine risk it poses. A calculator did one narrow thing, and you could spot-check it easily; a general-purpose AI drafts essays, builds arguments, and delivers fluent, confident prose that is harder to eyeball for errors. The wider the task you hand off, the more you needed to have done it yourself to judge the result. Recent studies—a 2025 Microsoft–Carnegie Mellon survey and a separate 666-person study on cognitive offloading—suggest that people who trust AI most tend to do less critical thinking themselves. The author treats these as strong hints rather than settled proof, correctly noting they are correlational and self-reported, and resists overclaiming.

The practical insight is aimed at parents and professionals: flip the usual order. Instead of using AI to skip the work, use it to critique work you have already done. "Here's my answer; what did I miss?" builds judgment. "Give me the answer" replaces it. This habit scales: a six-year-old doing it by hand is sounding out a word; a teenager is sketching an argument before writing. The point is not to reject AI but to use it only after you have built the understanding to know when it is wrong.

No comments yet. Be the first to share your thoughts!