'AI Is a Tool' Post Hits 233 Points on Hacker News: Skill Is What Matters

The yyhh.org blog post titled "AI is a tool and using it well takes skill" was published on October 12, 2026, and collected 233 points and 209 comments on Hacker News. We don't have the full text of the post, only the title and the discussion metrics. So we will not retell the author's arguments. Still, the thesis and the reaction to it show where the central dispute over AI tools currently lies.

Why a banal thesis sparks an argument

The post has nearly equal points and comments: 233 to 209. On Hacker News, that is usually a sign of debate rather than quiet agreement. The phrase "results depend on the user's skill" sounds like a truism, but neither camp likes it. Skeptics read it as an excuse for weak models: if a tool requires skill, it must be half-baked. Enthusiasts see it as spoiling the story that you just write a prompt and everything works.

Both positions can be tested against data. There isn't much data yet, but there is some.

What the measurements show

The 2023 BCG and Harvard study involved 758 consultants. On tasks inside the "jagged frontier" of GPT-4's capabilities, they completed 12.2% more tasks and worked 25.1% faster, and their output quality was rated more than 40% higher. On a task outside that frontier, the group with AI got correct answers 19 percentage points less often than the group without it.

In 2025, the organization METR ran an experiment with 16 experienced open-source developers on 246 real tasks. With AI tools, they worked 19% slower. Before starting, they had expected a 24% speedup, and after finishing, they believed they had been 20% faster.

Both results support the thesis about skill. The key skill here is understanding where the tool helps and where it gets in the way, and measuring that rather than trusting your impressions.

Metrics of the material and context

• Source: Hacker News, with a link to the yyhh.org blog.
• Publication date: October 12, 2026.
• Post score: 233 points.
• Comments: 209.
• BCG/Harvard (2023): +12.2% tasks, 25.1% faster, quality more than 40% higher inside the capability frontier; 19 percentage points fewer correct answers outside it.
• METR (2025): 16 developers, 246 tasks, 19% slower while expecting a 24% speedup.

Where to start in practice

Skill at working with AI can be broken down into a few concrete habits.

• Provide context. The model doesn't know your code, your constraints, or your definition of "done" until you describe them.
• Break tasks into pieces. A short step with a clear result is easier to verify than a huge draft.
• Verify results with tests, linters, and cross-checks against sources, not by eyeballing them.
• Decide in advance which tasks you won't hand to the model at all: those where the cost of an error is high and checking costs more than doing the work.
• Measure. For two weeks, record how long typical tasks take with and without AI. The METR result shows that your internal sense of speed is easily fooled.

A separate note for managers: buying licenses does not create skill. If the team is untrained and has no metrics of its own, you can end up with tool costs and a negative effect on speed. It's best to check the original post's text for the specific advice the author gives. The Hacker News discussion at the link also offers many practical examples from people who use these tools every day.