Do you know how to write complete registration/login validation logic?
I saw someone in a group sharing a ridiculous vibe coding failure case. Someone used AI to write a client-side application and had AI generate a piece of registration/login validation logic. During testing, they discovered that if you entered a password string that someone else had used during registration, the system would prompt: "This password was already used by ×××× during registration. Please use a different password to register."
The consequence of this approach is that you could randomly make up a password, and if you guessed correctly, you could steal someone else's account.
This reminded me that more and more people in the industry are vibe coding nowadays—handwriting code seems to have become some kind of retro activity. Whether it's login systems, APIs, backends, reports, or customer service bots, everything is handed over to AI to complete. But whether what it writes actually works still depends on whether the person using it understands the business.
In this case, if the AI-delivered project were handed over to someone who completely doesn't understand the business for testing, they would likely find it difficult to perceive the consequences of this erroneous interaction. AI only knows how to write closed-loop business logic; it doesn't know whether such closed loops make sense. So when the person controlling the AI also doesn't understand, disaster arrives as expected.
Many companies are currently undergoing AI transformation—they've paid consulting fees, attended training courses, and burned through tokens, yet achieved no results in the end. Their problem lies here too: they think that as long as they connect various AI tools, tinker with a knowledge base, and conduct a round of AI tool usage training, they can achieve so-called AI transformation.
This is purely wishful thinking. AI can reduce implementation difficulty, but it won't lower business barriers. If you don't understand, you don't understand—no amount of AI will change that. The following 7 pitfalls are the easiest traps for traditional enterprises to fall into when undergoing AI transformation.
Treating AI transformation as a "decision problem"
When most business owners hear a new buzzword, their first reaction is to spend money to ask an "expert" about it, and then have the "expert" develop a plan.
If this were a mature market, this approach would work. Because the market has already formed effective methodologies, "experts" can directly copy others' plans—even if they don't perfectly match, they can help bosses avoid some detours.
But the AI sector is different. This is a brand-new technology, and most people are still exploring methods for implementation. There are few methods that can be directly applied. Asking "experts" will only get you explanations of terminology that may not be correct, and asking them for plans is like asking a blind person for directions.
Before AI transformation, you need to ask yourself: why are you undergoing AI transformation?
Confusing AI strategy with AI transformation
At a public meeting about strategy a few years ago, Pony Ma complained that many strategy experts are very good at making beautiful strategic plans, but in actual business operations, strategy is made rather than thought out.
No wonder Pony Ma had such complaints—internet professionals are the group least likely to believe in "strategy" stuff. For people close to the tech frontlines, no amount of strategy is as good as market validation facing users. Because in the tech industry, everything changes too fast. By the time the strategic plan comes out, every word written in that document is probably already outdated.
When facing unfamiliar territory, no matter how beautiful or exquisite the strategy is, it's not as effective as practical implementation.
Treating training as transformation
Many companies think that as long as they have employees attend round after round of prompt training, agent usage training, and AI thinking training, they've completed AI transformation.
But this is just skills training. Employees learning to use some efficiency tools will only make the company's original business more efficient, but it won't change the original form of the business, nor will it make the business "AI-ified." Not to mention that most of these training skills are skills that will quickly become obsolete. How many of those "magical prompts" that were repeatedly shared a year ago are still effective today? Almost none, right?
Even Claude engineers claim that as AI capabilities become stronger, they are simplifying prompt usage more and more. In most scenarios, they only need one or two sentences to define an agent's capabilities.
Treating the procurement of AI tools as AI transformation
All bosses who think that deploying various AI tools in the company means completing AI transformation should understand one thing: tools are just one means of achieving goals.
Many people mistake means for ends, and "doing" for "doing it right." This gives rise to many strange operations done just for the sake of doing.
Like the DeepSeek all-in-one machines of 2025, or the Lobster Wave at the beginning of 2025.
Without even understanding the capability boundaries of a tool, let alone configuring the tool for your own business, you charge in blindly just because you heard that a certain tool "represents" productivity. In ancient times, this was called impulse buying or blind consumption.
Treating tool deployment as successful transformation is usually bound to employee training. "AI mentors" need a visible tool as a deliverable for training courses; otherwise, they can't complete the final delivery, affecting subsequent payments and secondary sales—this is the general routine of the industry.
But the problem is: Having a Ferrari doesn't mean you can run F1. Not to mention that most Ferraris you think you have are just Faralis (knock-offs).
Thinking that building a knowledge base makes you AI-ready
The knowledge base is a magical thing. Before AI, it was an overlooked business component that played a crucial role in consulting and training—back then, everyone would sing praises of McDonald's standardized knowledge system. After AI, it became something everyone flocked to—you're not AI without it, and with it, you don't know how to use it.
Those kings of the old era—their knowledge bases didn't just contain clearly defined and concisely summarized business methodologies; they were the code of conduct for the enterprise.
The knowledge base itself is just a technical tool for document storage and indexing. What really makes it valuable is the knowledge it contains. Not all private enterprise information deserves to be called "knowledge"; for most enterprises, it's just Garbage in, garbage out.
From a technical perspective, many knowledge bases deployed by enterprises use outdated versions of RAG technology, capable only of outputting specious fragmented information without being able to express causal relationships in business logic.
From a business perspective, a knowledge base cannot solve the problem of an enterprise having no knowledge.
Replacing human customer service with AI customer service as completion of AI transformation
Binding a Q&A knowledge base and replacing human customer service on corporate websites, internal assistants, and business windows with AI chatbots seems to reduce labor costs and alleviate customer service pressure, but in reality, it distances the enterprise from its customers.
In the old logic, the essential purpose of customer service for large enterprises wasn't to solve customer problems, but to control costs and filter demands. Because these enterprises had market dominance, external customer service was only meant to solve low-value queries and basic Q&A, without bearing brand and sales targets—and chatbots happen to be very good at this business.
But for most enterprises with insufficient brand power and channels, they need customer service to bear the responsibility of sales conversion, secondary sales, and closing deals. Handing this over to chatbots is like drinking poison to quench thirst. They neither have the capability to prevent prompt injection attacks, nor can they make chatbots complete customer guidance through simple system prompts.
For SMEs, sincerity is the ultimate weapon. Real human face-to-face interaction is the greatest sincerity.
Treating AI replacement of employees as the end goal of transformation
Many company bosses get excited the moment they hear AI can replace employees, as if they've found a shortcut to ascend to heaven by stepping on their own left foot with their right.
The wave of Lobster Brains (naive AI adopters) created by the Lobster Wave at the beginning of the year thought AI could automate everything. But as mentioned at the beginning of the article, AI is good at writing closed-loop business logic, but not good at judging which closed loops have value.
Moreover, if most of a company's business can be taken over by AI, it precisely shows that this company has no innovation capability and no market competitiveness.
Using AI to take over highly repetitive work is the correct approach. The human resources freed up should be concentrated on technological innovation and new business development—this is where the value of AI replacing human labor lies.
First do things right, then think about how to do things well. If replacing human labor stops at replacement rather than expansion, you're treating the starting point of transformation as the endpoint—which looks awkward and fatal however you look at it.
The real question of AI transformation needs to return to a sentence mentioned earlier: What are you undergoing AI transformation for?
Transformation is a means, not an end. Your end comes from your motivation. Is it for cost reduction and efficiency improvement, or because you've encountered business bottlenecks? Or is it simply FOMO—feeling that everyone around you is doing AI, and you'll seem out of place if you don't?
Business decisions should remain as rational as possible. Everyone selling AI concepts will try to tear open your rationality.
Any action should be based on specific business goals. AI won't make enterprises smarter; it will only make them expose their true level faster.