Many people believe that AI programming is a liberation for developers, but based on my experience with vibe coding over the past year, this claim is debatable.
Recently, I made an application for a friend and planned to supplement it with a user manual in the past couple of days. While creating the manual, I discovered a minor bug and took the opportunity to do a round of optimizations. Unknowingly, a whole day and night passed.
The most laborious part during this time wasn't writing prompts, nor was it brainstorming ideas; it was testing. It was comprehensive testing from morning till night, from top to bottom, from left to right. Changing a tiny interaction required me to retest all possible paths because I knew there would inevitably be unexpected issues. And indeed, that was the case.
I have this kind of "awareness" purely from stepping into too many pitfalls. Forget whack-a-mole; I've even encountered situations where tweaking some UI ultimately led to an entire business process being deleted. So, every time during the final delivery phase of software, I have no choice but to test again and again.
This "verification anxiety" stems from the fact that vibe coding cannot guarantee that each output will effectively converge, because the nature of LLMs dictates that they will still do things outside the intended constraints.
Traditional applications are deterministic systems; what cannot be constrained is user behavior. As long as the business process is correct, the results are bound to meet expectations. But LLMs are different; they are non-deterministic systems based on natural language. And natural language inherently contains ambiguity. The same word, at different times and places, can mean different things to me and to the model.
For over a year now, many self-media outlets have been touting that AI will replace programming, that coding is cheap. Indeed, the cost of writing code has significantly decreased, but in contrast, the cost of verification is evolving into a heavy burden. The inversion of development cost and testing cost will redefine business processes.
To ensure that these probabilistically generated codes don't crash in some environment, I must pay a hefty "trust cost." For an application that was just casually talked into existence, even if I have "strategically" directed my various AI "experts" to implement sufficient defenses (whether the AI actually did it or not, I don't know, but I've issued the commands), I know there will be many pitfalls waiting for me. To prevent the application from crashing in front of users, I have no choice but to crash in front of the application first.
The software industry is still a relatively easy field to converge and verify, after all, code is a form of expression with relatively strong determinism. But this migration of trust cost will certainly spread to all walks of life as AI becomes more widespread.
Once this "cost migration" leaves software and enters the real world, the problems will become extremely weighty. For example, in fields like healthcare, law, or embodied intelligence, you can't manage things with Git like in programming, there's no undo, and "hot updates" are even more impossible. AI can provide a seemingly extremely professional diagnosis suggestion or legal document in one second, but the effort people spend to identify the hidden, potentially fatal logical deviations within it might be more than if they had written one themselves.
The technical debt generated by AI will inundate every field that aims to "use AI to improve quality and efficiency." When generating functionality becomes almost free, ensuring that the AI's output is "trustworthy" instead becomes the most expensive luxury of this era.
The future will be a long-tail era with infinitely extended trust costs. People's past experience and intuition gained from stepping into pitfalls will become the hardest assets.
AI will lower the entry barriers for all industries to an extremely low level, but it will also make the ceiling unreachably high. Business processes will shift from "solving problems" to a new paradigm of "proving that the problem has been solved." The judgment to identify all kinds of deep pitfalls, big and small, within the vast ocean of AI-generated content, is the truly priceless wealth a person can possess.