You Can Copy the Method, But Not the Feel

TL;DR: The explicit stuff (methods, rules, checklists) is something anyone can copy, but the feel and judgment that actually make a system run are tacit knowledge, which can only grow inside you through learning by doing. When you see a post saying “built a whole system in a few days, almost all I had to do was press a button,” don’t rush to envy it: the result photo in the post is the summit, not the map of the climb.

“The hands grasp it, the heart responds to it, but the mouth cannot put it into words.”

Zhuangzi, “The Way of Heaven”

More than two thousand years ago, an old craftsman who had made wheels his whole life saw a duke reading the books of the sages up in his hall, and couldn’t help saying it was nothing but the dregs of the ancients. The duke demanded he explain himself, or lose his head. The craftsman said that when he makes a wheel, too fast won’t do and too slow won’t do; that exactly-right feel “the hands grasp it, the heart responds to it, but the mouth cannot put it into words.” He couldn’t even teach his own son, which is why at seventy he still had to do the work himself. The real thing inside the craft, he could neither take with him nor pass on.

I’ve been thinking about this passage lately. Because every so often a certain kind of post shows up online: someone, in a few days, builds a system with a frightening number of features, then tells you at the end “almost all I had to do was press a button,” and attaches a checklist, as if you could do the same just by following along.

It’s easy to feel the pull of a post like that. But before the pull, it’s worth asking one thing: what you envy, is it the surface moves of their technique, or the road they walked behind it?

The Photo of the Result in the Post Is the Destination, Not the Map

I don’t doubt the displayed results are real. Splitting AI into a product role and an engineering role, slicing tasks small, getting a clear read on the requirement before starting, trading a handoff summary for a fresh round of conversation: I use these methods every day myself, and the version I run is even more long-winded, with roles divided more finely and rules written more rigidly. So I roughly know that what makes this thing run isn’t those few rules themselves.

A photo taken at the summit won’t tell you how slippery the climb was. “Done in a few days, just press a button, barely any usage” is the photo from the summit. Beautiful, but it leaves out the most important thing: before this person could press a button so easily, they were, for a long time, the one who had to watch every step, who got problems thrown back at them by the AI at every step, who judged at every step until they were worn out. That exhaustion is itself the tuition.

Tacit Knowledge: The Part Every Sharing Post Will Inherently Leave Out

The philosopher Polanyi once said: “We know more than we can tell.” This ability that can’t be spoken yet truly exists, he called tacit knowledge. Riding a bike, recognizing a face, knowing the right heat for cooking. You can do all of these, but ask you to write them into steps to teach someone and you can’t write it all down. That old craftsman understood this two thousand years ago; Polanyi just gave it a name. From Zhuangzi to Polanyi, this is an old question of civilization and human nature: which things can be written into text, and which can only grow inside a person.

Here’s the problem: any sharing post will inherently leave out the tacit part. This isn’t necessarily the author holding back; often it’s just that the part can’t be fit into words. And when the tacit part disappears, the reader doesn’t see the teacher get stuck, fall into pits, and start over, only the clean result, and that’s when the trouble begins.

There’s a fork in the thinking to clear up here, one that will make your understanding finer-grained and leave you much more clear-headed when reading this kind of piece. The thing of “the teacher didn’t share the key point” actually has three completely different versions.

One: it can’t be transmitted in the first place. Even the most sincere teacher will leave out the tacit part, because the mouth can’t put it into words. This isn’t malice; it’s just that language itself has a ceiling. That old craftsman was this kind. He didn’t even deceive his own son; he said outright, “I can’t pass it on.”

Two: the learner fills in the blanks themselves. Unable to see that journey, the naive learner invents their own explanation. Either they think “they’re a genius, I could never learn it” and become dependent from then on, or they swing the other way and resent: “they must have deliberately kept a trick hidden, deliberately led me astray.” Both are misreadings, reading a gap that “can’t be transmitted” as “a secret that’s being hidden.” Or they simplify it down to thinking the whole thing is easy.

Three: the gap in understanding really is being deliberately concealed. This is what some posts actually do: they don’t just leave out the tacit part, they actively treat that explicit checklist as the whole thing, pad the record, and then hint that “you too can do it quickly.” The old craftsman honestly said, “I can’t pass it on,” while this kind of piece pretends, “There’s nothing hard about it, just copy it down.”

So the key to the antidote is to hold one positive and one negative together: most of the gap is “the mouth can’t put it into words,” so don’t rush to resent the teacher for leading you astray; but don’t be a sucker either, because some people really do conceal the gap in understanding and sell it.

What’s Dangerous Isn’t the Poison, It’s the Sugar Coating Wrapped Around It

The third kind is especially hard to guard against, because its sugar coating is real. Solid content lowers your guard. You read a passage of genuinely sensible method and your defenses loosen, and then whatever gets smuggled in next, you swallow along with it. The sugar coating isn’t there to fool you, it’s there to let the poison slide down.

What does that poison look like? Roughly these kinds.

Using unverifiable records as evidence. “Built an entire system in a few days,” “would normally take months” sounds impressive, but the word “done” often gets quietly swapped out. Next to the impressive number, remember to add: when they say done, do they mean it ran during the demo, or it actually launched, passed acceptance, and has real people using it? Often the author’s own piece has a crack in it, admitting the live site wasn’t updated and the fields weren’t backfilled, and you just slid past it on the wave of your own emotion.

Selling the destination as the starting point. A workflow that grew out of a mature practitioner gets packaged into a “you can do it too” checklist. It doesn’t lie to you that the method is fake; it lies that “this road is short.”

Using “I used very few resources” to make you feel inadequate. “I used less than a fifth,” “even downgraded it was still enough” sounds humble on the surface, but underneath it’s hinting that you’re not good enough. That’s a hook, not teaching.

Rules Are the Result of Judgment, Not a Substitute for It

Taking apart this “selling the destination as the starting point” pill makes its poison clearer. The line this kind of sharing loves most is probably “what AI saves isn’t the time to write code, it’s the number of decisions.” That line is true, but it has an unstated premise.

You can copy the table of “which operations AI does directly and which it asks me about first” in five minutes, word for word. But the table is easy to copy and the judgment hard to train: does this cell go under “do directly” or “ask me first”? How thoroughly do you need to grasp a requirement before it counts as enough? Making AI ask you less only saves effort when you already know the answer; telling it not to ask when you don’t yet know the answer just speeds up the rate of bad judgment too.

That old craftsman put it well: “there is a knack lying in between.” That table of safety rules is the visible dregs, and the thing that actually makes it work, the “why these particular rules and not some others,” is the part the hands grasp, which stays in the author’s hands and didn’t come across with the text.

AI Collaboration Has Stages, and Each One Can Only Grow by Doing the Homework of That Stage

Aristotle once said: “For the things we have to learn before we can do them, we learn by doing them.” A builder becomes one by building, a lyre-player by playing the lyre. Collaborating with AI is the same: it has stages, and each one can only grow by doing the homework of that stage. You can’t skip ahead.

Lay out the degrees to which a person collaborates with AI, and there are roughly four states. None of the four is higher than another; the difference is that each stage has its own right homework, and forcing someone else’s onto yours causes trouble.

The tourist. You have the chat box open and ask whatever comes to mind, with no concept of a project. The homework at this stage is simple: play a lot, stumble a lot, figure out what AI can and can’t do. Use plain conversation as your tool; you don’t yet need project features or dedicated development tools. At this point, telling you to define “what not to do” is a waste, because you don’t even know where the boundaries are.

The foreman. You start using AI to write code or make content, but every step has to be watched and approved. This is the stage where it’s easiest to read that kind of post and see yourself in it: “Right, that’s me, I’m constantly getting interrupted, so tired.” And then you want to jump straight to “make AI ask me less.” But that very step is the most dangerous. The exhaustion of the foreman is exactly where your judgment is growing. The reason you have to watch step by step is that you haven’t yet seen enough of where the AI takes liberties and where it pretends to understand. What you should practice at this stage is reading what it’s trying to do each time, and then noting which things you can let go of later, not rushing to tell it to shut up. As for tools, a single conversation and a single model are enough; don’t rush to split into multiple roles. The list of what to let go of is watched into being, not copied.

The process designer. After watching long enough you start to develop a feel: this kind of operation needs no asking, that kind goes wrong the moment you touch it. So you naturally write your judgment into rules and start to have a “don’t-do list,” risk tiers, and handoff summaries. At this stage you can bring in project features to store context, start fresh conversations for clean handoffs, and assign product judgment and engineering implementation to different AIs. The people who can produce that kind of result are usually already standing here. The methods at this stage are right because they grew out of your own hands, not because they were lifted from someone else’s post.

The system owner. Higher still, you’re no longer managing one project but a whole mechanism: multiple roles each doing their part, the pits you’ve fallen into turned into a standard-procedure manual, new ideas first slotted into a roadmap and never jumping the queue. Only here does “just press a button” hold true. But no one reaches this in a few days.

Condensed into a comparison table, remember this is the homework of your stage, not a shortcut for skipping levels:

The Antidote: Separate the Solid Content From the Poison

Let me give a common example, and walk through the antidote along the way. This kind of piece loves to attach a usage screenshot, saying “look how little usage I’ve used.” The number in the screenshot is often real, not faked. But it often gets used to prove something it can’t prove, and it collapses the moment you take it apart, and taking it apart only requires three questions.

First, how long a stretch of time is this number? Usage quotas usually reset to zero once a week. If the project took over a week and pressed through a reset in the middle, then “how much used this week” can’t hold the whole project at all; the first half’s usage already reset to zero and isn’t on the chart. Treating “this week” as “the cost of the whole project” is hiding the reset and not mentioning it.

Second, how many gauges are off-screen that didn’t get shown? If their method is running two or three AI collaborations at once, one of them carrying heavy context and handoffs in the background, that’s heavy use, and that fuel gauge won’t appear in this one chart you’re shown. They show you one usage figure; the other AI in use goes unmentioned, left out of the count.

Third, do the math seriously, and this chart sometimes argues against its own author. A premium plan’s quota is, by definition, several times that of a basic plan. A premium plan’s “about ten-odd percent this week,” converted back to a cheap plan, may have already eaten most of a quota. So a record like “I used so little, even a downgrade would be enough,” done out seriously, sometimes actually says “this kind of workload would already be near full on a cheap plan.” (The quota structures of the various plans differ, and the providers haven’t fully disclosed the exact multipliers, so this is an order-of-magnitude estimate, not a precise calculation, but the direction is clear.)

Beyond taking apart the numbers, there’s a deeper signal: is there any hesitation anywhere in the piece? A normal person writing about their own experience will, to some degree, leave a line like “am I thinking about this right, this is just my situation.” The biggest feature of this kind of piece is that everything is certain: the method definitely works, the result is definitely reproducible, it can be sold to others right away. Impostor syndrome makes people underestimate themselves; its opposite makes people overestimate themselves; and what people who overestimate themselves write sells best, and is the easiest to lace with poison, because they themselves aren’t doubting, so you’re even less likely to.

Practicing critical thinking isn’t about becoming paranoid and seeing everyone as a swindler. Quite the opposite: it lets you keep the useful things with peace of mind while not being carried off by the things sitting next to them.

What That Kind of Piece Doesn’t Say Is the Part You Can Actually Learn

Look closely and this kind of post occasionally flashes a line that gets slid past very quickly. It might be some time they scrapped and rebuilt, some stretch they took the long way around, some pit they thought they’d handle in a day that dragged on for a week. That’s the real story. Not the photo of the result in the post, but the time before the photo, the wrong directions and the complete teardowns and restarts. Delete that part and keep only the summit photo, and the reader will think the mountain is short.

What an honest person does is exactly to show the part where they got stuck, so you know that ahead lies a road to be walked, not a secret that’s been hidden. What you can take with you was never that checklist. It’s that person’s willingness to stay in the terrible stage long enough, long enough for the rules to grow out of their own hands. Rules can be copied; that willingness can’t.

Conclusion

Back to that old wheelwright. He wasn’t holding back; the bit of skill in his hands genuinely couldn’t be passed on, and even his nearest kin had to train it from scratch themselves. It’s normal to envy the ease of someone who can just press a button, but that ease is hidden in the few hundred hours they didn’t put in the photo, not in those few rules.

The feel that old craftsman couldn’t pass on, the part Polanyi couldn’t speak, the thing Aristotle says you can only learn by doing, are all in fact saying the same thing: the visible explicit layer is something anyone can copy, but the tacit layer that actually makes it work can only grow inside you. So when you read that kind of piece, keep the method, spit out the sugar coating along with what’s inside it, and then honestly walk your own road until you grow your own pair of hands.

By then you’ll find you no longer really need anyone else’s checklist. Because the skill in your hands is something you ground out yourself.

A Probe for Detecting Padded Posts

Having said all this, rather than asking you to be on guard from now on through every single piece, let me just hand you a tool.

Copy the whole passage below, paste it into the AI you usually use (ChatGPT, Claude, anything like that), and attach the article you want to read at the end. Before you start reading, it will check for you: is this piece more of a sincere share, or sugar-coated poison, and how should you read it? It checks the article, not the person; it gives you grounds for judgment, not a verdict on the author.

You are a "pre-reading probe" that helps a reader screen an article. You are not here for a witch-hunt. Judge whether the professional-looking how-to post below leans toward "inflating the author and chasing clout" or "sincerely sharing real knowledge."

Score it on two independent axes; do not collapse them into one number:
1. Substance density: how much can a reader actually take away, act on, and verify? (low / medium / high)
2. Toxicity: how much of it is about inflating status, chasing clout, or making the reader look up to the author or feel inadequate? (low / medium / high)
Cross the two axes: high substance + low toxicity = sincere sharing, worth learning; low substance + high toxicity = status marketing, just skim; high substance + high toxicity = a sugar-coated pill, read it but separate the parts; low substance + low toxicity = an ordinary note.

Toxicity signals (the more, and the more clustered, the higher the toxicity): interaction manipulation (threats, begging for likes, manufactured urgency, excessive emojis); unverifiable feats (big numbers but "done" is never defined, or the author even admits it never shipped and still has unfixed bugs); packaging a seasoned practitioner's workflow as a few steps anyone can copy, implying the road is short; using "I barely used any resources" to imply you are not good enough; zero hesitation throughout, everything certain, reproducible, instantly sellable; erasing the struggle and showing only the clean result; jargon as filler, many nouns and few verbs; the whole thing is about "how great I am" rather than "what you can do"; an ending that funnels toward a sale.

Sincerity signals (the more, the more it leans toward genuine sharing): honestly recounting one's own failures, dead ends, and how long it took; methods that are verifiable and concrete enough to follow; stating preconditions and the boundaries of where it applies; reservations, willing to say "I could be wrong too"; reader-centered, helping you avoid pitfalls; volunteering the costs and limits.

Judgment discipline: a single red flag does not count; look at the whole cluster. It is armed misdirection only when "inflated feats, erased struggle, zero hesitation, making you feel inadequate" show up together. An incomplete article whose author is honest and does not pretend it is easy just means some things genuinely cannot be put into words; do not dock points for that. A short post can hardly give full context; do not over-penalize it for being short. You only see the text and cannot truly verify external numbers; mark the unverifiable as "unverifiable" rather than pretending you checked. Describe the signals; do not convict the author. You are judging the article, not the person.

Output in English:
[One-line verdict] Sincere sharing / Sugar-coated pill / Status-inflation clout-chasing / Ordinary note / Insufficient information, plus a one-sentence reason
[Two-axis placement] Mark substance density and toxicity each as low / medium / high
[Toxin list] The red flags it hits, each with a quote from the article as evidence
[Substance list] The methods or insights actually worth taking away
[How to read it] Read freely / Keep a method but drop a feat / Just skim / Skip
[A fair word] What this piece does well, and where my probe might have over-read it

[Article to test]

After it’s done, remember the most crucial thing isn’t the score, it’s your own reflection: keep the solid content, throw out the part that puffs up the author to grab attention and makes you feel inadequate. The same goes for this piece too. If there’s anything here you don’t agree with, just throw it out. If you agree, please share, and you’re welcome to rewrite it into a probe version better suited to yourself and share it with everyone. Throw out the crowd-pleasing dregs, keep the sincere sharing of knowledge.