A blog post titled "AI is a tool and using it well takes skill" climbed the Hacker News front page this week, collecting 287 points and 238 comments. Its author, writing as Yyhh, takes aim at Rich Hickey's claim that LLMs are automation rather than tools. The takeaway for you: whether AI makes your work better or just faster and sloppier depends largely on what you bring to it.
This piece borrows that argument and turns it into practical habits you can apply whether you write code, research, or learn new subjects.
Why "tool versus automation" is more than semantics
Per the essay, the author argues against Hickey's framing and makes the case that AI can be wielded skillfully when guided by human judgment, feedback, and oversight. The distinction matters because of what each word implies about your role.
Automation suggests you press a button and walk away. A tool suggests you hold it, steer it, and answer for what it produces. If you accept the first framing, quality is somebody else's problem. If you accept the second, quality is yours.
You don't have to settle the philosophical debate to take the practical point: the label you choose changes how much attention you pay.
AI amplifies what you already do
The essay frames LLMs as instruments that amplify the user's habits, standards, and expertise rather than replacing them outright. That's the most useful idea in it, and it cuts both ways.
If you have high standards and know your field, the model can help you draft, test, and refine faster. If you have loose standards or shaky understanding, the same tool will happily amplify that too. (That second half is my inference from the amplification idea, not a claim lifted from the essay.)
So the first skill is honest self-assessment: where is your judgment strong enough to evaluate the output, and where aren't you sure you'd notice a mistake?
Practical habits for using AI without outsourcing your thinking
The essay's broader theme is guidance, feedback, and oversight. Here is how I'd translate that into routines.
Decide first, then ask
Before prompting, write down what you think the answer or approach is, even in two sentences. Then compare. You'll see where the model adds something and where it just agrees with you or drifts.
Ask for critique, not just output
Instead of "write this," try handing over your own draft, code, or argument and asking where it's weakest. This keeps you as the author and the model as a reviewer, which supports learning, critique, and iteration rather than replacement.
Iterate with feedback
Treat the first response as a rough pass. Tell the model what's wrong, what's missing, and what constraints matter. Skilled use looks like a loop, not a single request.
Verify what you can't judge
If you can't tell whether a claim, function, or citation is correct, that's your cue to check it elsewhere before relying on it. Accountability stays with you.
A workflow for programmers
- Write the spec or test cases yourself. Defining what "correct" means is the part worth keeping.
- Let the model propose an implementation.
- Read every line as if a junior colleague wrote it. Ask it to explain anything you don't follow.
- Run the tests, then ask the model to critique its own solution for edge cases.
- Make the final call yourself and own the commit.
A workflow for researchers and learners
- Form your own question and a tentative answer first.
- Use the model to surface counterarguments or terms you haven't encountered.
- Follow up in primary sources before you trust a summary.
- Explain the idea back in your own words. If you can't, you haven't learned it yet.
Don't treat it like a person
One of the angles the essay's framing points toward is the danger of anthropomorphizing AI. When you treat a model as a colleague with intentions, it's easy to defer to its confident tone. Treating it as an instrument keeps the question simple: is this output good, and how do I know?
Between hype and rejection
As AI shows up in writing, coding, research, and education, the debate is shifting from "should we use it?" to "what counts as responsible use?" The essay pushes back on both hype and rejection by focusing on skill, agency, and accountability. That's a sensible middle path: you neither hand over the wheel nor refuse the vehicle.
The one thing to do this week
Pick one task you routinely give to an AI tool and add a single step that keeps your judgment in the loop: write your own answer first, or ask for a critique of your draft instead of a replacement. Then notice whether your results, and your understanding, improve.
