Vibe Coding vs Real Programming: Why We Teach Both
Updated: 2026-08-15 | Reading time: 5 min
There are two loud camps. One says AI tools make learning fundamentals pointless. The other says vibe coding produces people who cannot actually program. Both are describing a real thing, and both conclusions are wrong.
The case the AI-first camp gets right
Refusing to teach these tools would be indefensible. Surveys in early 2026 suggest the overwhelming majority of professional developers now use AI assistance in some form. A student who graduates having never used it is entering the job market with a handicap we gave them.
They are also right that some traditional teaching wasted enormous time. Weeks of syntax drills made sense when a misplaced semicolon could cost an evening. It costs seconds now.
The case the traditionalists get right
A student who cannot read code cannot tell working software from software that merely runs. They cannot debug, cannot judge whether a suggestion is sensible, and cannot be trusted near anything involving other people's money or data.
The research backs the worry: learners who lean on AI generators show weaker retention and markedly weaker debugging skill. That is not conservatism, it is a measured effect.
What we concluded
These are not opposing philosophies to pick between. They are two skills that reinforce each other, so we run them in the same weeks rather than in sequence.
Every week has two halves. In one, students learn a Python concept properly — writing it by hand, breaking it, explaining it. In the other, they use Claude Code to build something real that uses that same concept.
The pairing does something neither half achieves alone. The fundamentals give students the judgement to evaluate what the AI produced. The AI project gives the fundamentals a visible purpose, which is what keeps a fourteen-year-old turning up on a Saturday.
What "both" looks like in one week
Take the week on dictionaries:
- Fundamentals half: build a small phonebook by hand. Add, look up, delete. Break it deliberately and fix it.
- Project half: use Claude Code to extend it into a real tool — save to a file, search partial names, print a clean list.
- The bridge: the student explains what the AI changed and why, and finds at least one thing they would have done differently.
That last item is the whole course in one line.
The honest limitation
This is slower than pure vibe coding. A student who only prompts will have a more impressive-looking app in week three than one of ours. By week twelve, when something breaks in front of a client, the comparison reverses — and that is the trade we have chosen deliberately.