UpTrajectory Review

A piece by Diomidis Spinellis, surfaced on Hacker News with 222 points and 288 comments, takes on the question of what happens to human decision-making when AI tools become the default first stop for thinking through problems. The title frames thinking as a 'hobby' — a deliberate, almost recreational act — in contrast to the instrumental, outsourced cognition that AI increasingly absorbs. Spinellis is a software engineer and author of long standing in the systems and tooling community, so the argument arrives with credibility among technical operators, the same audience that has adopted AI coding assistants and reasoning tools fastest. The HN traction alone signals that this tension — delegation versus judgment — is live for a lot of people right now.

For a small-business operator, this is not an abstract philosophy question. If you run a lean operation, you are already using AI to draft emails, model scenarios, review contracts, or generate marketing copy. The risk Spinellis is pointing at is not that AI gives you bad answers — it is that you stop generating your own questions. When thinking becomes something you only do when the tools are down, your ability to spot when the tool is wrong, when the framing is off, or when the problem itself is misdefined atrophies quietly. Operators who never developed the muscle of sitting with a problem before reaching for a solution are the most exposed when AI confidently produces a polished, wrong answer.

What is genuinely contested here is whether this is a new problem or an old one with new plumbing. Critics on HN will reasonably point out that operators have been outsourcing thinking to consultants, spreadsheets, and frameworks for decades — AI is just faster and cheaper. That is fair, but it misses a real difference in degree that becomes a difference in kind: AI does not just execute your thinking, it simulates the thinking itself, fluently and without friction. The under-reported angle is what this does to organizational memory. When the reasoning behind a decision lives in a prompt history rather than in a human's worked-through logic, the business loses the ability to audit, teach, or improve that reasoning over time.

The second-order effects split the operator audience in two. Experienced operators with deep domain knowledge will likely use AI to accelerate thinking they already know how to do — a genuine productivity gain. Less experienced operators, or those in newer domains, risk building their entire decision-making process on top of a tool they cannot evaluate, which compounds quietly until a high-stakes decision goes wrong in a way that is hard to diagnose. There is also a cost dimension: the time saved on individual decisions may be repaid in full the first time a bad AI-influenced call costs you a client, a hire, or a contract. The economics only work if the operator stays in the loop.

What to watch is whether tools start building in friction deliberately — forcing users to articulate their own reasoning before seeing AI output — and whether any credible training or certification emerges around AI-assisted judgment for non-technical operators. In the meantime, the practical move is simple: before you prompt, write down your own answer or hypothesis first. Treat AI as a sparring partner, not a substitute. If you cannot articulate what you would do without the tool, you are not ready to use it. That discipline costs almost nothing and preserves exactly the capability Spinellis is warning we are giving up.

The HN comment thread, at nearly 300 comments, is worth reading alongside the piece — it is likely where the sharpest counterarguments and real-world operator experiences live. The piece itself is a provocation more than a prescription, but that is precisely its value. It names something most operators feel but have not articulated: the creeping delegation of the one skill that justifies their role. Whether you find the argument persuasive or overblown, it is the right question to sit with before your next big call.

Takeaway: Write your own answer before you prompt — AI should pressure-test your thinking, not replace it.