Human in the Loop

Are You Ready for Vibe Coding? Yes, You Are!

• By Howard M. Cohen
• August 20, 2025

When you first put your hands on your first low-code/no-code (LCNC) platform and began creating applications by moving those tiles around the screen, you probably had no expectations. You may recall the pleasant surprise you felt when you ran your first LCNC application and saw it doing what you intended it to do. Surprising, right?

Prepare to be surprised again!

You Ain’t Seen Nuthin’ Yet
You’ve undoubtedly heard or seen the phrase "vibe coding," and you may even be worried that this is something you’re not ready for. The good news is you’re wrong about that.

If your earliest results with LCNC knocked you out, you’re probably going to be blown away by what you create with vibe coding, and how ready you really are to get started… today!

Quick History
Andrej Karpathy is one of those true experts you want to follow to learn more about artificial intelligence (AI). Many of the videos on his YouTube channel are meant for non-technical audiences, and he has a talent for achieving clarity in his explanations of complex issues. His session on "Software in the Age of AI" at the Y Combinator AI Startup School is nearly legendary. After a stint as a research scientist and a cofounding member of OpenAI, Karpathy became Senior Director of AI at Tesla, after which he soon returned to OpenAI and most recently founded AI+Education company Eureka Labs.

Karpathy initially introduced vibe coding in February 2025. He defined it as using large language models (LLMs) to generate code from plain-language prompts, rather than writing code manually. It's characterized by a conversational workflow: you describe what you want, the AI generates the code, and you review or tweak as needed.

You're Already More Ready Than You Know
You’re probably already working with an AI large language model (LLM), like ChatGPT, Gemini, Claude, Perplexity, DeepSeek, or Grok. For consistency, I’m going to refer to each as a "model." Although I don’t recommend any model in particular, many analysts have pointed to Claude from Anthropic as being very good for vibe coding. However, many of them recognized the recently released ChatGPT 5 as being somewhat better when used in "think hard" mode. You’ll find your preferred model as you play with each of them.

It's a Conversation
Most citizen developers who try vibe coding find it easier than they ever thought. They find themselves simply conversing with the AI model they’re working with. It seems natural to them. As citizen developers, they're focused on solving specific problems. Vibe coding involves describing problems and intended software solutions in plain language, providing as many detailed instructions as possible in the prompt.

We recently discussed prompt engineering and context engineering, and there are many training materials, courses, tutorials, and other guidance available. Still, you’re going to find that vibe coding doesn’t really require much formal training. The materials make great suggestions that may improve your results on the first try.

The good news is that you don’t have to stop at the first try. If you don’t get exactly the result you were trying for, just keep conversing with the model in the prompt window. It will work with you to get closer and closer to your desired result until you are there.

Vibe Coding Builds Better Problem-Solving Skills
We often emphasize that the best way to use AI is to treat it as a partner, and excellent partnerships work both ways. Unlike following platform tutorials, vibe coding encourages you to think through problems more thoroughly. When something breaks or doesn’t work as intended, you work with the model to figure out why, which makes you a stronger application developer overall.

This is really no different from "real" application development. Professionals who code spend a lot of time experimenting, reading Stack Overflow, and trying approaches that "feel right." What you’re learning from vibe coding is sometimes called "transferable intuition."

The First Thing You Must Prepare is Your Mind
Messiness and dead ends are part of the process, not signs you're doing it wrong, so go easy on yourself. Like riding a bike or any other skill, you will make mistakes, though you’re not likely to skin your knees. The goal in vibe coding isn't to write perfect code.

That’s why it refers to "vibes." You’re working with your AI partner to solve problems effectively while gradually building your intuition, your insight, and your sense of what works and what doesn’t. The transition to vibe coding is less about changing what you do and more about giving yourself permission to trust your instincts. Relax. Enjoy it.

That Doesn’t Mean Throw Structure to the Winds
You don't want to abandon structure entirely; even vibe coding benefits from your ongoing reflection on what's working and what isn't. Don't let experimentation become an excuse for not learning fundamentals when you hit real limitations. Those learning materials, and the model itself, provide plenty of ways to figure your way past obstacles.

Don’t be afraid just to ask the model how you could do a better job of prompting it. It will tell you.

How Will Vibe Coding Differ from Low-Code/No-Code?
The good news is that citizen developers find several key differences between vibe coding and typical low-code/no-code platforms that help them do a better job faster and generate more capable applications than ever before. These include:

• Freedom: Most low-code platforms guide you down predetermined paths, essentially filling in templates or connecting pre-built components. Vibe coding is far more freeform, encouraging you to break out of those guidelines, mixing tools unconventionally, or letting the model write custom code for you when the model’s platform limitations feel restrictive.
• Discovery: Low-code platforms often walk you through step-by-step processes, imposing their own structure. Vibe coding skips the structure and lets you quickly and easily hack together prototypes using different tools to see which way works best.
• Choose Your Tools: Low-code solutions typically want you to stay within their ecosystem. Vibe coding, like all of AI, embraces whatever works. You might use Airtable for data storage, a Python script for processing, and a simple HTML page for display, even if it's "messier" than using one integrated platform. Or you might prefer Outsystems Mentor to do the entire job. You get to choose which and how many tools are involved based on experience with them.
• Experiments: Low-code platforms tend to emphasize professional-looking results that follow strict design patterns. Vibe coding discards all that and replaces it with a simple, "Does this work for me?" These experiments usually go into production but remain experiments… forever!
• Play: Even "easy" low-code platforms have concepts to master and best practices to follow. Vibe coding says "try stuff and see what happens" without requiring you to understand the platform's mental model first. You’re having a conversation with a partner. Play around with it. Have fun together making work better, easier, and more productive.

The difference is one of philosophy. Low-code platforms aim to make software development more accessible by introducing structure, whereas vibe coding achieves this by lowering technical barriers, eliminating the need for structure and syntax, and embracing experimentation in everyday language, thereby offering a new freedom to explore, innovate, and create.

Get Started with Vibe Coding
Stick to these simple suggestions and you’ll find yourself vibe coding before dinnertime. 

1. Start with What You Already Know – Pick a fairly simple real-world pain point you’re confronting, preferably something repetitive, manual, or otherwise "clunky" like automating the generation of some reports, or building a simple dashboard to keep track of rapidly changing data. Perhaps just some internal form used for requesting benefits, PTO, or office supplies. The more you care about the problem, the more you’ll stick to figuring out how vibe coding will solve it. about.
2. Choose the Right Vibe-Coding-Friendly Tools – There are some good LCNC platforms, like the aforementioned Outsystems, or the Microsoft Power Platform, that already integrate LLM-powered coding assistants. There are also a growing number of general AI models specifically designed for coding, including GitHub Copilot Chat, Replit Ghostwriter, or ChatGPT’s code interpreter (Advanced Data Analysis). Try to give yourself time working with these before trying others, or you could end up hopping from one to another endlessly.
3. Learning to "Speak" to the AI is probably not new to you. You’ve already done that with digital assistants like Alexa, Siri, or Google Assistant. It’s easier for you to learn how they want to hear it than to try to get them to recognize the way you want to say it. In the case of vibe coding, you’ll find it works best when you write prompts like mini-specifications with clearly stated goals, as many key details as you can think of, and as many step-by-step building instructions as possible, knowing you can always go back and add more if needed. Also, remember these are computers that tend to take things very literally.
4. Short Loop Interactions, where your Prompt → AI Output → Test → Refine will serve you best. Don’t try to get it perfect in one go—citizen developers excel when they prototype fast and keep the conversation going. Just remember to keep a "what worked" prompt log so you can reuse good instructions later.
5. Build a Light Foundation in Core Concepts because even without deep coding it helps to learn data types (text, number, date), understand basic logic (if-then, loops), get familiar with application programming interfaces (API) as "connectors" between tools and learn as much as you can about how to read simple code so you can tweak AI output without starting over.
6. Join one of the many Vibe-Coding Communities that are springing up all over, like the Microsoft Power Platform Community, the Replit Community Hub, the NoCodeDevs forums, or Slack groups, so you can share what you’re building. The feedback you get from these groups will sharpen both your prompting and your solution design.
7. Work Toward Autonomy - Over time, you’ll move from "copying" AI output to customizing it, learning when to trust AI and when to sanity-check results, and experiment with more complex builds—multi-screen apps, data pipelines, automations.

Remember, you’re not learning to code in the traditional sense—you’re learning to collaborate with AI to translate ideas into working solutions. The skill is design thinking + clear communication, not memorizing syntax. And that’s good because syntax is complicated.

It’s Alright to Get it Wrong
With so many models to choose from, numerous tools emerging to help, and constant advancements happening at a rapid pace, there really is no such thing as "right" or "wrong" in this new world of vibe coding. So, pick your first model knowing it may not be the best for your purposes. Explore tools that may not be useful to you, and ask your model to create your first application, even if it doesn’t work as expected. Still, you can then just tell the model why it doesn’t work and quickly go from somewhat wrong to kinda right to wow, that’s just what this citizen developer ordered!