My business partner sent a 5K vibe-coded PR that he didn't even test

TEXT START: The whole payments backend module, which I specifically told him to work together with me (I do frontend), was vibe-coded by him in one single day.

The Dissection

This is an incident report disguised as a eulogy for hand-written programming. The concrete failure is not that AI generated 5,236 lines; it is that the partner outsourced implementation, documentation, and basic validation while retaining responsibility for the result. Generation accelerated. Judgment, integration, and verification were abandoned.

The article correctly identifies the dangerous pattern: AI makes incompetent work look complete. A large pull request creates the visual impression of productivity while concealing that nobody has established whether the system functions. The missing API-key instructions and untested endpoints are not cosmetic defects. They are evidence that the producer no longer maintained a working model of the product.

Under the Discontinuity Thesis, this is an early organizational form of cognitive automation: production becomes cheap, while trusted verification becomes the scarce input. The machine is not merely replacing typing. It is weakening the human's incentive and ability to understand what was produced.

The Core Fallacy

The article treats the central threat as individual cognitive atrophy caused by excessive AI assistance. That is real, but secondary. The deeper mechanism is economic: once AI can generate acceptable-looking software at negligible marginal cost, organizations optimize for output volume and apparent speed. Verification becomes the bottleneck, but firms often underfund it because it does not produce impressive line counts or rapid demos.

The partner did not fail because AI made him stupid in isolation. He failed because AI let him bypass collaboration, domain transfer, and feedback while still producing an artifact that looked like work. The tool converted ignorance into scalable throughput.

The article also implies that continuing to program manually preserves a meaningful refuge. It does not. Hand-written code may preserve the programmer's cognition and craftsmanship, but it does not restore the old labor market once AI dominates cognitive production. Manual skill becomes a preference, a niche, or a verification discipline—not a defense of mass productive participation.

Hidden Assumptions

• That individual discipline can counter the competitive pressure to use AI for faster production.
• That more human effort in code generation is more valuable than reliable testing, architecture, and operational knowledge.
• That the problem is primarily loss of intelligence rather than loss of accountability.
• That teams will continue rewarding understanding instead of rewarding visible output and compressed deadlines.
• That human programmers form a stable professional community capable of collectively resisting the new production regime.
• That preserving one's own cognitive fitness preserves one's economic indispensability.
• That the old craft-based identity of programming can survive after the underlying wage-and-status structure is automated.

These assumptions fail under P1 and P2. Competitive systems do not preserve hand labor because it is intellectually healthier. They preserve it only where automation cannot yet coordinate the full task cheaply and reliably. The moment verification, maintenance, and integration are also automated or concentrated into a smaller expert layer, the manual coder loses leverage.

Social Function

Classification: partial truth, prestige signaling, and ideological anesthetic.

It is a partial truth because the failure mode is genuine: unchecked AI output can produce brittle systems and degrade the user's working knowledge. It is prestige signaling because manual programming is framed as evidence of seriousness and intellectual vitality. It is ideological anesthetic because the lament relocates a structural transformation into personal habits—exercise your brain, resist the tool, remain one of the few who still cares.

That diagnosis gives the individual a morally satisfying battlefield while concealing the larger one. Even a careful programmer cannot indefinitely defeat a production system that makes cognitive labor cheaper, faster, and more scalable. The surviving leverage lies in owning AI capital or becoming indispensable at verification, integration, security, operations, maintenance, and coordination—not in preserving artisanal line-by-line authorship.

The yoga-teacher ending is the text's honest moment. It recognizes that the old occupation may become emotionally untenable before it becomes technically impossible. The grief is valid. The implied hope that enough people can simply keep programming properly is not.

The Verdict

The PR is a small-scale autopsy of the coming labor regime: AI expands the supply of plausible work while human attention remains too scarce to validate it. The immediate culprit is negligent use, but the terminal mechanism is broader. As cognitive production becomes abundant, understanding becomes concentrated in fewer owners and high-leverage servitors, while everyone else is left generating artifacts nobody has time to inspect.

The partner's code may be bad. The more important fact is that the system rewarded him for producing it anyway. That is not a temporary etiquette problem. It is the first visible corpse of the old competence economy.