When I published Vibe Coding Is Not a Strategy, Scott Pfister left the best comment. “It’s not vibe coding,” he wrote. “It’s vibe shipping.”
He’s right. And it sharpens the argument.
The generation method isn’t the crisis. AI writing your code isn’t inherently reckless. The crisis is what happens after the code is written: when it ships without structure, standards, or quality checks. That’s vibe shipping. And that’s what actually breaks teams.
The speed is real. The problem is what runs under it.
AI-assisted development is genuinely fast. A developer who knows what they’re doing and has good tools can ship features at a rate that wasn’t possible two years ago. That’s the promise, and for a lot of teams, it’s real.
But speed without a model around it doesn’t produce better software. It produces more software, faster, with the same underlying problems. Sometimes worse ones. John S. put it well in the comments: “The risk is not just lower-quality code; it is faster drift from the system’s intended design.”
Drift is the key word. When you’re moving fast without a model, you’re not moving in a straight line. You’re scattering.
What the operating model looks like
Standards, written down before the AI touches the code. The AI will follow the patterns you give it. If you haven’t defined your patterns, it invents them. Christopher James Smith described it in the comments: “I’m not just telling it what I want, I’m telling it the pattern I want it to use, what edge cases to address.” That’s standards in practice.
Context loaded into every session. The AI has no memory of your system unless you give it one. Sean Michael Kelly named what that means: standards, guidelines, NFRs, and controls, all available to the model before it writes a line. Teams that skip this are asking the AI to guess at constraints it can’t see.
Gates on what ships. Code review still matters. Testing still matters. The fact that AI wrote the code doesn’t change what needs to happen before it reaches production. If anything, the volume makes the gates more important, not less.
A clear line between prototype and production. Krzysztof Dudek is right that a PM building a UX sandbox has different needs than an engineer shipping to customers. The mistake is when the mindset from one bleeds into the other.
Vibe coding is a style. Vibe shipping is a failure mode.
The problem isn’t that developers are using AI. The problem is that some teams adopted a powerful tool without asking what that tool needs around it to work safely at scale.
That’s not an AI problem. It’s an operating model problem. And it’s the kind of thing that looks fine at low volume and falls apart when the team gets bigger, the codebase gets deeper, and the speed that felt like a superpower starts compounding errors instead of features.
The challenge
Pick one AI coding workflow your team is using regularly. Write down the standard it should follow. Not a ban, not a lecture: just one pattern, documented, that the AI can reference and that your team agrees to enforce.
One standard. This week.