Not a Coder? With A.I., Just Having an Idea Can Be Enough.

I’m not a coder. I can’t write a single line of Python, JavaScript, or C++. Aside from a brief phase in my teenage years when I created websites and played around with Flash animations, I’ve never been a software engineer, nor do I have plans to shift my career from journalism to tech.

But over the past few months, I’ve found myself coding more than ever.

Among the things I’ve built: a tool that transcribes and summarizes long podcasts, a platform that organizes my social media bookmarks into a searchable database, a website that tells me if a piece of furniture will fit in my car’s trunk, and an app called LunchBox Buddy, which analyzes what’s in my fridge and helps me decide what to pack for my son’s school lunch.

These projects have been made possible by artificial intelligence and a growing trend in A.I. called “vibecoding.”

Coined by A.I. researcher Andrej Karpathy, vibecoding refers to how today’s A.I. tools allow even non-programmers to create fully functional apps and websites simply by typing prompts into a text box. You don’t need coding skills to vibecode—just a clear idea and a bit of patience.

“It’s not really coding,” Karpathy said this month. “I just see stuff, say stuff, run stuff, and copy-paste stuff, and it mostly works.”

My own vibecoding efforts have focused on what I call “software for one” — small, customized apps that address personal needs. These aren’t tools big tech companies would create. There’s no market for them, their features are simple, and some only work partially.

But the process of building software this way — describing a problem in a sentence, then watching a powerful A.I. model craft a solution — is an awe-inspiring experience. It’s like A.I. vertigo, similar to how I felt the first time I used ChatGPT. It’s also a powerful way to show skeptics how much today’s A.I. models can do, especially when they automate key parts of computer programming. In the future, they may be capable of similar feats in other domains.

A.I. coding tools aren’t new. Earlier versions, like GitHub Copilot, were designed to assist professional programmers by completing lines of code, much like ChatGPT finishes sentences. You still needed coding knowledge to get the most out of them and step in when the A.I. stumbled.

But in the last year or two, tools that leverage more advanced A.I. models have emerged, allowing even beginners to create like seasoned developers.

These tools — including Cursor, Replit, Bolt, and Lovable — function similarly. When given a prompt, they design the project, choose the appropriate software packages and programming languages, and start building. Most of these tools offer limited free use, with paid options unlocking additional features and more building capacity.

To someone with no coding experience, vibecoding feels like magic. After typing your prompt, lines of code appear seemingly out of nowhere, and within minutes (or hours, depending on complexity), a working prototype materializes. Users can suggest changes, and when satisfied, deploy their creation to the web or run it on their device.

Here’s an example: when I asked Bolt to help me build an app that suggests lunch ideas for my son based on a photo of my fridge, it went through a series of steps. First, it analyzed the task and broke it down. Then it generated a basic interface, chose an image recognition tool to identify food items in my fridge, and created an algorithm to recommend meals.

If the A.I. needed clarification — like whether to list nutritional information — it asked me for input. When it hit a problem, it either debugged itself or reverted to a previous step and tried a new approach.

Around 10 minutes after I submitted the prompt, my app, LunchBox Buddy, was ready. It suggested a basic turkey sandwich. You can try it out here. (Note: I replaced the A.I. image recognition tool with a mock version to avoid a hefty fee for this public demo.)

Not all my vibecoding projects have gone smoothly. I’ve been struggling to build an “inbox autopilot” that responds to emails in my writing style. I’ve run into issues trying to integrate A.I. workflows with apps like Google Photos and iOS Voice Memos, which weren’t built for third-party add-ons.

And of course, A.I. isn’t perfect. Once, when trying to create a website for a local tire shop, the A.I. generated fake Yelp reviews and added them to a testimonial page. Another time, it chopped up a long story I wrote into a website, leaving half the content out.

In short, vibecoding still benefits from human oversight, or at least supervision. It’s better suited for hobby projects than critical tasks.

That may not last long. Many A.I. companies are developing software engineering agents that could fully replace human programmers. Already, A.I. has scored world-class results on competitive programming exams, and big tech companies like Google are using A.I. for a significant portion of their engineering work. Sundar Pichai, Google’s CEO, recently revealed that A.I.-generated code makes up over a quarter of new code deployed at the company.

If I were a junior programmer, I’d be worried about my job prospects. But I’m just someone who enjoys tinkering and creating tools that make my life a little easier. And in that regard, vibecoding — or even full coding — is undeniably improving with A.I.

Since sharing my vibecoding experience on my podcast, I’ve heard from many people who are also using A.I. to build their own tools. Some have developed apps to help with dieting or summarize newsletters, while others have built websites to track egg prices or uncover rent-gouging in Los Angeles after the Palisades fire.

None of these tools will change the world, but what’s remarkable is that with just a few keystrokes, amateurs can now create products that once required entire teams of engineers.

I’m not naïve about A.I. or blind to the potential societal impacts if A.I. coding tools continue to improve. There’s a chance these tools could also automate the creation of harmful code, or lead to autonomous cyberattacks. I’m also concerned that software engineering might just be the first white-collar field to face A.I.-driven job displacement.

But for now, using A.I. to build apps that automate small, tedious tasks in my life seems like a pretty good use of this technology. So I’ll keep vibecoding — at least until my kid can pack his own lunch.