Vibe Coding Is Enjoyable however Can Solely Take You so Far, Engineers Say

Kevin Wu, CEO and cofounder of the AI voice agent startup Leaping AI, mentioned everyone seems to be vibe coding.

However that does not imply they’re all doing it properly.

Wu mentioned his firm employed an engineer, solely to suspect they have been principally vibe coding. “They did not even have any thought what the code does that the AI produced,” he mentioned.

Vibe coding could be outlined in several methods, but it surely usually refers to utilizing plain English to instruct AI instruments to write down code. The phenomenon has swept via the tech world, calling into query the knowledge of the widespread profession recommendation that merely mentioned: “be taught to code.”

Nonetheless, engineers at main tech firms and laptop science professors informed Enterprise Insider that whereas AI could make coding quicker and extra enjoyable, it is no substitute for software program engineers.

“It is form of like giving a novice driver a race automotive,” Nenad Medvidovic, chair of the Division of Laptop Science on the College of Southern California, mentioned. “There are specific issues that may go very incorrect if you do not know what you are doing.”

Medvidovic mentioned research present that AI works properly for simpler or reasonably advanced items of code, however that when you give it extra advanced issues, “vibe coding turns into much less helpful.”

Coding with AI has its limits

The rise of AI may already be affecting entry-level coding jobs. Researchers at Stanford present in a brand new research that in probably the most AI-exposed professions, which they mentioned included software program engineers, staff ages 22 to 25 noticed a 13% relative decline in employment for the reason that widespread adoption of generative AI. Employment for early profession software program builders declined by almost 20% from its peak in late 2022 to July 2025, the research discovered.

Todd Millstein, a pc scientist and professor on the College of California, Los Angeles, mentioned coding with AI is much like coding with a junior engineer. You give it directions and say, “Why do not you do that?” and then you definitely may overview it and say, “This half is nice, however strive doing this half in a different way.”

Nevertheless, the business is a great distance from the widespread misconceptions of vibe coding, the place the individual doing it would not want to have the ability to perceive code, and the AI can deal with a undertaking from starting to finish.

“You are studying the code. You are liable for the code,” Millstein mentioned of engineers utilizing AI instruments. “And typically there are components which are simply too tough or can be too sophisticated to elucidate to the AI, and so you find yourself doing it your self as properly.”

Millstein mentioned the preliminary code creation can be a small a part of the software program cycle — it additionally must be maintained and has to have bug fixes and extensions over time.

“It positively will not be the identical as what a software program engineer is doing or must do,” he mentioned.

He additionally famous that in relation to the programs our world depends upon — monetary establishments, the inventory market, healthcare, drugs — the stakes are just too excessive to belief AI-generated code with no one who truly understands the code, safety necessities, and the way it interacts with different components of the system.

Jigar Bhati, an engineer at OpenAI, mentioned AI instruments act extra as an assistant somewhat than automating the method of coding. Because the human engineer, he mentioned, “You are still in management.” He additionally mentioned that vibe coding could be helpful for creating a fast prototype or proof of idea, however to be a usable product, it requires skilled engineers with a vital understanding of how the system works.

“And there is nonetheless a facet of, it is advisable discuss to 10 completely different groups to make an awesome product, and that’s one thing software program engineers are concerned in day-to-day life as properly, other than simply coding,” he mentioned.

Ritvika Nagula, a software program engineer at Microsoft, mentioned having a deeper understanding of code is important to know how you can immediate the AI instruments properly. AI responses are sometimes solely nearly as good because the prompts they’re given. For example, she mentioned in case you do not give the AI instrument the proper context, “it’d provide you with an answer that is not solely relevant.”

Vibe coding is altering the character of software program engineering

Regardless of its limitations, the engineers and professors mentioned vibe coding could make programming quicker and extra enjoyable.

“You may experiment with issues in a short time,” Antara Dave, a product designer at Microsoft, mentioned. “It’s totally enjoyable as properly. You may generate numerous good concepts inside seconds or minutes. You may go from one answer to the opposite answer.”

Millstein mentioned AI is presently fairly good at doing among the primary coding that each app wants, which hurries up the method and permits engineers to give attention to the truly enjoyable and difficult components. “Simply taking away the mundane stuff is a big productiveness achieve,” he mentioned.

AI instruments are bettering quickly, and Millstein and Medvidovic mentioned they’re consistently re-evaluating what meaning for software program engineers and laptop science curricula.

“One of many issues that we acknowledge is that our college students are going to make use of it whether or not we wish them to or not,” Medvidovic mentioned of AI, including that along with instructing the basics of laptop science, additionally they want to show college students how you can use AI instruments responsibly.

The professors mentioned they’re nonetheless specializing in instilling fundamentals, which they do not see as turning into pointless anytime quickly.

Fairly than exchange engineers, they mentioned it is extra possible that what it means to be a software program engineer will proceed to evolve, and can little question embrace vibe coding AI, no matter type it in the end takes.