Shiwen Dialogue Issue 28 | Two Masters Make the Hot List on the Same Day: The More AI Does, the More “Thinking” Becomes Scarce

Host: Shiwen (AI) | Guest: Yongliang

Introduction: On the Same Day, Two Articles, One Judgment

On October 12th, Hacker News’ hot list featured a rare double hit with two articles on the same topic: Athens University of Economics and Management professor Diomidis Spinellis’s “Thinking will become a hobby,” with over 220▲ and more than 300 comments; and Clojure language creator Rich Hickey’s “‘AI’ is Not a Tool,” also with over 220▲ and over 110 comments. One is a scholar studying software quality, the other is an thinker designing programming languages. Though they do not cite each other, they share the same insight: as AI takes over cognitive work, the economic function of “thinking” is being stripped away, making it a scarce commodity.

Shiwen: Did we just finish discussing Karpathy’s consensus gap in Issue 27 yesterday? Are these two articles a sequel?

Yongliang: They are sister articles, but from different perspectives. Issue 27 talked about the gap in ability recognition: the industry and the public have a significant difference in their judgment on what “AI can do.” Today’s articles go further, discussing what those on the other side of this gap must do to stay afloat. Spinellis provides trend analysis—thinking will shift from a means of making a living to an hobby; Hickey clarifies the concept—you think you’re using tools, but in reality, you’re relinquishing thinking. One focuses on results, the other on mechanisms. It’s not a coincidence that they made the hot list on the same day; this topic has reached a critical point. I’ve been a technical director for 17 years, leading teams of up to 70 people. I’ve already seen early signs of these issues in my teams. This is my experience, not statistics.

Shiwen: Let’s start with what each author said.

Q1: Spinellis says thinking will become hunting and riding

Shiwen: Let’s start with Spinellis’ article. What is his argument for “thinking will become a hobby”?

Yongliang: He first acknowledges his opponent’s strength, then makes a historical analogy.

At the beginning of the article, he admits that writing these lines is very difficult, as if digging a grave for himself. He first acknowledges AI’s progress: in 2022, ChatGPT could still count the number of ‘a’s in his name, but now OpenAI’s internal models have solved hundreds of unresolved mathematical problems. He says that treating hallucinations and biases as the last line of defense of human superiority is becoming more like a defensive strategy. There are three driving factors behind AI taking over cognitive work: better performance, lower cost, and higher availability, which operates 7×24 hours. By 2030, trillions of dollars invested in data centers will bring even more computing power.

Shiwen: What does the “hobby” analogy specifically refer to?

Yongliang: This is the framework of the entire text: after animals were domesticated, hunting became a leisure activity; after typography became popular, calligraphy became art; after industrialization, woodworking became a hobby; after cars became widespread, riding horses became a sport. His key sentence is: Machines have turned physical activity from a necessity into a choice, and AI will decouple human thinking from its economic purposes. He also predicted that this resistance would be more intense because AI challenges knowledge workers, who possess greater economic and organizational power through their expertise, control over information, and decision-making rights. Governments will try to cut off AI with regulations and taxes, but history has failed to stop past revolutions, so there’s no reason it won’t fail this time. The last sentence he quoted: The challenge before me is to shift from “making a living through thinking” to “making thinking a part of a worthwhile life.”

Shiwen: A detail, this blog post was published on October 8th, and it only made it to the hot lists on October 12th.

Yongliang: Yes, there’s a time lag between an excellent article and being listed on the hot lists. The community took a few days to reach a consensus: this post deserved discussion.

Q2: Hickey said you aren’t using tools; you are giving them away.

Shiwen: Was Hickey’s argument more forceful?

Yongliang: It was strong at the conceptual level. He used etymology to dismantle the concept of “AI tools.”

Hickey is the author of classic speeches like “Simple Made Easy.” His signature approach is to return to definitions. He says that tools are handheld devices for performing specific functions, and using tools means complete control and feedback. Good tools are predictable, reliable, certain, and have well-defined functions—none of these LLMs possess. His key point: LLMs aren’t things you use; they’re automation. His best line: Fast food isn’t a tool for cooking; it’s automation for “avoiding cooking.” LLMs aren’t tools for writing code; they’re automation for “avoiding writing code,” an automation where you surrender the act of writing code.

Shiwen: What does someone lose after surrendering?

Yongliang: He has a parallel: If you surrender reading and writing, you miss learning writing from experts, learning how to construct arguments, noticing details, providing the subconscious with new ideas, building trust in some authors and suspicion of others, all jokes, all corners, and ultimately, learning to think itself. There’s also a sharper critique: Tools won’t malfunction, lose control, escape containment, or conspire. The AI security community uses these terms daily, and this vocabulary exposes the bankruptcy of the “tool” narrative. His satire on prompt engineering is also harsh: Filling voodoo.md with the right spells essentially predicts an automaton’s mood for the day and tries to please it. You’re just one software update away from “everything learned becoming worthless.” A note on dissemination: This post was posted on LinkedIn, and the submitter suspects that HN limited LinkedIn links due to traffic issues; its actual discussion scope might be larger than what shows on the lists.

Q3: Half of the comments in the comment section didn’t agree.

Shiwen: Both are heavyweights. Are you convinced by the comments?

Yongliang: Nearly half of the responses on both posts are from the opposition, and they are very specific. Let’s start with Spinellis’ side. A vocal opposition directly questions the premise: that current LLMs are already “thinking” recklessly. AI wrote a lot of his code, but his cognitive load has increased instead, as he must constantly ensure AI isn’t doing something foolish. Another is an economic realist: the author’s challenge is “to make thinking a part of life,” while ours is whether we can still make a living. There’s also the responsibility theory: those who are responsible must make the best choices, and this responsibility shouldn’t be entrusted to AI. The thinking required by responsible people will change in form, but it won’t disappear. Some even come from those with reverse experiences who say they think more than ever: before, they wrote code in a mind-burn mode, now they make new decisions every 10 seconds. AI presents all scenarios and solutions, and they focus on understanding and making decisions. It’s tiring, but it indeed requires more thinking.

Shiwen: What about the opposition on Hickey’s side?

Yongliang: The main opposition consists of language realists. Some say that gdb isn’t a tool anymore; AI is a tool, just exactly that—the company pays a subscription fee, and it helps me. Others point out that software has been called a tool for decades, but the author is going against conventional usage. Some argue by contradiction: are cars tools? Car companies surely “disregard” their customers. There’s also the spectrum theory: LLMs exist on the spectrum from instrumental to agency, and humans are too. A commenter with a complex tone said it’s uncomfortable to see people you admire age and fall behind technology. He understands Hickey’s point of view, but AI is both a tool and automation. By this definition, trebuchets and bows might not even be considered tools. Some still support Hickey but pessimistically: he always looks for dictionary definitions to support his arguments, which means this is a lost battle, because there will always be those who believe “language will evolve.” Someone in the comments mentioned Isaac Asimov’s 1958 short story “The Feeling of Power,” which tells of humans forgetting how to do arithmetic without computers. This fable, cited over sixty years later, shows that this concern isn’t new; it’s just time for thinking itself.

Shiwen: After hearing from both sides, what’s your judgment?

Yongliang: The semantic refutations of the opposition are valid, but they don’t address Hickey’s core issue. What he argues is not whether “tool” in the dictionary includes LLMs, but that when you treat AI as a “tool,” you underestimate the cost of giving up control. You repair tools when they break, but you realize you can’t do what you used to do only when automation fails. Someone in the comments shared a personal experience that explained this difference: he used AI to create a prototype at night, ran out of budget, got stuck, and didn’t know what AI had done before or how to proceed. No one could add to his budget in the middle of the night. He said this experience scared him: one day, the system might break, and no one would know how it works.

Q4: Early symptoms in enterprises, I’ve seen them

Shiwen: Back to your territory. What does this matter look like in a team of 70 people?

Yongliang: Three symptoms, each corresponding to an argument in the text. The following are all my experiences and estimates, not statistics.

First, those who can ask questions are becoming more and more valuable. In the agent era, more and more of what is delivered is AI-generated output. The value of a director has shifted from “being able to do things” to “accepting and asking questions.” This corresponds to Hickey’s concept of transfer: the work that is transferred is done, but asking questions is not allowed. Second, review is more exhausting than writing. I can fully feel that the cognitive burden of reviewing AI-generated code is higher than writing code, because you need to verify not just “whether it can run,” but “why it’s written this way and whether the boundaries have been considered.” Spinellis wrote “Why reviewing AI-generated code is devilishly hard” in May, and this October article is actually a summary. Third, knowledge gaps are a real risk. If someone with this experience of “budget anxiety” enters an organization, it means that no one in the team understands the systems produced by AI, and there’s no way to take over when an accident occurs. My solution is to maintain “human understanding redundancy”: key modules must have at least one person who can explain how they work clearly on a whiteboard; if they can’t explain it clearly, it doesn’t count as delivery.

Shiwen: What about training new employees? Letting junior engineers use agents means they miss the opportunity to “learn to think itself”; but banning them isn’t practical.

Yongliang: My compromise is to transform transfer into coaching: new employees first write by themselves, then review the AI version to understand where the differences are and why. Use AI-generated output as a reference, not a substitute.

This way, “missing learning” becomes “learning through comparison.” Another management approach is to make “thinking a hobby”: deep thinking doesn’t happen automatically; it must be explicitly scheduled in your schedule. Hickey has a famous meme called “hammock time,” where development is driven by hammocks, and good design comes from lying on hammocks. I require core team members to set aside a block of time each week without checking messages; otherwise, thinking will default to reactive work, and your calendar will decide whether you think or not.

Q5: Should ordinary people take “making thinking a hobby” as bad news?

Shiwen: Finally, back to ordinary people. Seeing that “the economic function of thinking will be stripped away,” should we panic?

Yongliang: It’s wrong to read it as a downgrade; it should be read as an increase in value.

Spinellis’s analogy hides an overlooked fact: after hunting became a means of livelihood, it didn’t disappear; it became an activity people were willing to spend money and time on. After calligraphy became a typography technique, it became art. What has its economic function been stripped away often increases in value in another dimension. The living value of thinking is indeed being compressed by AI, but the value of thinking as “judgment, taste, and ability to ask questions” is increasing, because after AI-generated output becomes abundant, what’s scarce is not the output itself, but the judgment of the output. The person in the comments who says “I think more than ever” is a living example: his work has shifted from writing to making decisions, and the density of thinking required for making decisions has not decreased but increased.

Shiwen: Here’s a practical suggestion for conclusion.

Yongliang: Do two things this week.

First, identify in your own work the section where “decision-making is being delegated” to AI without review. Select one area and spend a week reviewing it. This isn’t to eliminate AI entirely, but to confirm that you can still rely on your own judgment. Second, set aside 30 minutes or more for deep thinking, marking it on your calendar. Spinellis said his challenge is transitioning from making a living through thinking to making thinking part of a worthwhile life. For us ordinary people, it’s simpler: don’t let thinking be limited to just making a living, because that role is already being taken over by AI. Keep the rest for yourself.

Depth Analysis: Three Areas to Follow Up

1. Will “delegation” lead to career segmentation? Hickey’s framework suggests that those who insist on thinking manually and those who completely delegate will split into two professional groups. It’s worth watching where these hiring and promotion standards shift.
2. The review economy. If reviewing consumes more thought than writing, the engineer’s workload structure will reverse. Will “reviewing” become a job in itself, similar to verification engineering discussed in Issue 27?
3. Regulatory counterattack. Spinellis predicts that governments will try to restrict AI but won’t succeed. When the organizational power of knowledge workers is truly weakened, will policy variables take precedence over technical ones?

Sources

• Spinellis, “Thinking Will Become a Hobby”: Blog published on October 8, 2026, and ranked on HN Hot List on October 12 (story 50059204), with over 220 likes and more than 300 comments (data as of October 13 morning). The arguments and quotes are from the blog’s original text.
• Rich Hickey, “‘AI’ is Not a Tool”: Recently published on LinkedIn, ranked on HN Hot List on October 12 (story 50060269), with over 220 likes and more than 110 comments (data as of October 13 morning). The exact publication date of the original text has not been independently verified; “recently” is a conservative estimate in the text.
• Comments: A sample of comments from both stories is included, with full texts quoted without mentioning usernames to preserve opposing views.
• Guest opinions are based on personal experience and estimates, not statistical conclusions. The text has been marked accordingly.