Andrej Karpathy's "vibe coding" captured the initial excitement of 2025, but that term falls apart the moment a project hits a production environment and crashes because the "vibe" didn't account for edge-case memory leaks. For those of us doing professional work, saying "I let an agent write this" sounds less like a workflow and more like an excuse.
The problem is that brand-specific verbs—like "Clauding" or "Clankering"—feel like marketing speak, not engineering. I've tried a few alternatives to describe the daily grind of prompt engineering and agent orchestration, but nothing quite sticks:
Tokencrafting: Describes the precision of managing context windows, but feels too academic.Vibecrafting: A nod to the original term, but still lacks the "engineering" weight.Steering: Accurate, but sounds like I'm driving a boat rather than shipping a product.
In a real-world AI workflow, the actual labor has shifted from syntax mastery to architectural oversight and rigorous verification. We aren't typing characters; we are validating intent and debugging the output of a probabilistic engine.
If we aren't "writing" software anymore, the most honest description is probably that we've become editors of machine-generated logic. We spend 10% of our time prompting and 90% of our time ensuring the agent didn't hallucinate a library that doesn't exist. Until the industry settles on a new term, we're stuck in this linguistic limbo between being a "coder" and a "prompt operator."
[Next Vibecoding: The New Wave of App Store Chaos →](/en/threads/2741/)