AI Is a Skill Multiplier, Not a Skill Replacement

If you're getting inconsistent results from AI — output that looks right but keeps missing the mark, or a sense that colleagues are getting more out of the same tools than you are — the gap is almost certainly not the model. A piece by Josh Comeau, published May 22, reached 337 points on Hacker News without announcing a new model or a better prompt technique. It documented what actually separates people who compound with AI from people who plateau. The finding is uncomfortable and immediately useful.

What Comeau found

Comeau watched an expert developer use AI to roughly triple their output. Not because AI wrote code for them wholesale, but because they had the experience to give precise instructions, catch errors on the first pass, and course-correct with a single follow-up. They knew what "done" looked like before they started. The AI removed the mechanical work. The judgment stayed with the person.

In the same piece, he documented a contrasting pattern from Reddit. "Vibe coders" — people who learned to prompt before they learned the underlying craft — hit walls fast. One spent three hours prompting a model trying to resolve a problem they eventually fixed manually in 30 seconds, once they understood what the problem actually was. The AI returned confident answers throughout. None of them were right. The person lacked the domain knowledge to recognise that.

His framing: AI is "Iron Man's suit." The suit is real and powerful. Without Tony Stark inside it, it falls over. (Source: Josh Comeau, personal site, May 22, 2026; 337 HN points.)

The Monday morning implication

The most important question about your AI setup isn't which model to use or which prompt format produces better results. It's: what do I know well enough to supervise?

That's where the multiplication happens. AI amplifies what you bring to it. Strong domain knowledge, the ability to recognise a bad output, clear criteria for what good looks like — these make AI dramatically more useful. Gaps in underlying knowledge get returned to you faster, with more confidence, and harder to spot.

Investing in your craft and judgment isn't a backup plan for when AI disappoints you. It's what makes AI worth using.

A senior account manager who gets it immediately

A senior account manager at a 150-person financial services firm has seven years of client relationships behind her. She starts using AI to draft quarterly review prep materials and client update emails. Within two weeks she's producing better work in half the time.

Not because she's mastered any particular prompting technique. Because she can read the first paragraph of an AI draft and know instantly whether the tone is right for this client. She can identify what's missing from a summary before she sends it. She fixes in two sentences what would take a junior colleague three rounds of back-and-forth. Her existing skill is the active ingredient. The AI cleared space for her to apply it more often.

A marketing associate hitting a ceiling

A marketing associate four months into their first role at a 30-person e-commerce brand uses AI for most of what they produce — drafts, briefs, summaries, social copy. The output looks professional. Their manager sends it back most weeks. Not because the AI failed — because the associate doesn't yet have the experience to know what good looks like for this specific audience, this brand voice, this brief.

They're generating confident-looking work they can't evaluate. The AI is amplifying a gap they can't see yet.

This is not an argument against using AI early in a career. It's an argument for learning to supervise the output at the same time — which means building the underlying judgment to know when something is wrong, even when you can't immediately articulate why.

The return on skill investment is higher now than it's ever been — because every capability you build gets multiplied by the tools running alongside it.