Vibe Coders Are Missing the Hard Part

Añadido: 2026-05-30

[00:00:00]The new era of LLM's has lead [music] to an interesting change in the tech world.

[00:00:04]Hobbyists think they are developers because they can write code with their [music] eye. CEOs even think AI replaces them totally. Both see a machine writing code and assume that's the job, [music] but it actually never was. Writing code is only the tip of the iceberg and below are nine levels that decide whether your software survives >> [music] >> or burns. This is what everyone sees.

[00:00:25]You prompt, you get the code, it runs.

[00:00:27]The wipe coder calls it done. [music] But the engineer calls it the easy 10%, the syntax. But the job of a software developer is everything [music] underneath. Generating code is quite easy, but understanding it is another discipline. So the wipe coder pastes the error back into the chat and hopes. If it gets stuck, there is [music] no way of recovering. The engineer reads the stack trace and knows where to look. 80% of this job is reading code and not writing code. If you wipe coded before, you probably ask yourself, why does the AI break something every time it changes one thing? The answer is no structure.

[00:01:02]Everything is coupled to everything. A wipe coder will ask, how do I [music] build this? But an engineer will ask, where does this belong? That's the difference between a project that dies in three months and one that grows. I've read multiple articles where someone stated, "Oh dear, AI deleted my feature, deleted [music] my production database."

[00:01:19]Yeah, because you have got no safety net. You need Git, you need branches, you need commits, you need pull requests. That is not an optional extra.

[00:01:27]So the wipe coder hopes that the next prompt [music] won't break anything. But the engineer makes small, manageable, and traceable commits. Engineers not only build something, but they also take the correct requirements for an application. "Build me an app [music] for a booking appointment." That sounds simple as a prompt. But what about the double bookings? Do you want your customers to pay twice [music] for a single booking? They won't be so happy.

[00:01:47]What about cancellations, time zones, holidays? So the wipe coder ships the happy path and calls it a product. But the real engineer thinks about the edge cases. He extracts the correct business rules. An AI can implement a spec, >> [music] >> but it can't tell you if the spec is wrong. That's the engineer's job. What is also very important is the architecture of a system. Now, it's not about code. It's the decisions that shape [music] the whole system. The data model, which entities, which relationships. Do you want to build a monolith or a microservice? SQL or NoSQL? Sync or async? The white coder says, "The AI suggested it way." But the engineer knows every choice has got a trade-off. For performance, for security, for cost, and totally owns it.

[00:02:28]AI will show you options, [music] but the decision is yours. As engineers, we just don't build features, we also build safety nets around the features. So, if we add something new, [music] the tests should still be green. A white coder will test later, if ever.

[00:02:43]Another topic is how to get your code into production. [music] Your low-code demo is not a product yet. It needs deployments, it needs secrets, logging, monitoring, rollbacks, live database migrations. The white coder will say, "Oh, it runs on my machine." But an engineer knows it isn't done when it runs. It is done when it can be operated.

[00:03:01]Another topic is the security minefield.

[00:03:04]AI-generated code looks like a house with an open door. So, the white coder may ask, "Does it work?" But an engineer will ask, "What happens when someone tries to break it on purpose?" SQL injection, XSS, CSRF, no rate limits, authentication is also not authorization. And data privacy isn't something optional. That's defined by the law. Another important topic is especially for the business owner is the cost reality. Is your application built for 10 users?

[00:03:29]>> [music] >> But what happens then if 100,000 show up? Everything will collapse.

[00:03:34]But what if you build your application for 100,000 users, but only 10 [music] show up? Then the bill will eat your budget. As an engineer, you don't want to build the most impressive solution.

[00:03:44]You want to build the most sensible one.

[00:03:46]The next level is the ownership [music] layer. That layer isn't even technical.

[00:03:50]The white coder will build exactly what [music] he was asked. But a good engineer will tell the stakeholder that a new feature that [music] takes 2 weeks, if implemented the way suggested, will take 2 months of technical debt later. And that might be okay. So, we have to stand behind our decisions, [music] even the wrong ones. So, the TLDR of this video is wipe coding automates tip of the iceberg, and that's [music] good. But everything below the surface is what makes you a real engineer. AI can be a multiplier. The more of the iceberg you understand, [music] the more powerful AI becomes as a tool. So, my suggestion is start at the tip, but don't stay there.