Developer workflows have moved fast. In a few years we went from chat windows, to AI-enhanced IDEs, to environments where the assistant is the primary author and the human reviews. With predictions that AI will write the overwhelming majority of code, the interesting questions stop being about generation and start being about reliability, efficiency, and trust.
How fast the workflow moved
Every step of that progression made writing code cheaper. None of them made verifying code cheaper. That asymmetry is the whole story of the next few years.
Vibe coding, vibe debugging
"Vibe coding" — Andrej Karpathy's term for working in a mode where you effectively forget the code exists — is now common. Its natural consequence is vibe debugging: when the output misbehaves and nobody on the team has a mental model of what was written or why, you are reduced to guessing at a system you did not author.
The half nobody automated
This is where the series gets its name. Going from vibe to viable means treating measurement and validation as first-class steps, not afterthoughts — running the code, benchmarking it, checking it against real workloads, and only then calling it done. Generation gets you a draft. Evidence is what makes it shippable.
