Vibe Coding in 2026: I Tried Cursor, Replit, Bolt, Lovable, and V0. Here’s What Actually Ships.
Five tools, one simple project, and roughly £55 spent to learn that “anyone can build an app” is only half true.
The short version: I am not a developer. I tried building the same simple internal tool in five vibe coding platforms over one week. One produced something that actually works. Two produced things that look like they work. One produced something beautiful that does nothing. And one reminded me, within ten minutes, that I am not a developer. Total cost: roughly £55 in subscriptions and credits. Total functioning apps: one, sort of.
On a Thursday evening in early March, I typed “build me a client tracking tool” into a chat box and watched an AI write an entire application in front of me. It took about four minutes. The app had a login screen, a dashboard, a form for logging client notes, and a table that displayed everything in neat rows. It looked like something a real company would use.
I called my partner over to see it. She was politely impressed in the way that means she didn’t fully understand why I was excited but could tell I needed her to look at the screen.
Quick note before I go further: nobody pays me to write about any of this. No sponsors, no affiliates, no free accounts. I paid for every tool mentioned in this piece with my own money, and in one case I’m still annoyed about it.
The reason I was building anything at all is that “vibe coding” had become impossible to ignore. Collins Dictionary named it their Word of the Year. A quarter of the latest Y Combinator startup batch reportedly has codebases that are 95% AI generated. Every tech newsletter I read had someone claiming that the age of the nontechnical founder had arrived: just describe what you want, and AI builds it.
I wanted to know something simpler. Can a person who does not code use these tools to make something that actually works?
My qualifications for this experiment: I can write a VLOOKUP. I once edited a CSS file to change a button colour and it took forty minutes. That is the complete list.
The project
I kept this deliberately small. My team uses a shared spreadsheet to track client conversations: who we spoke to, when, what was discussed, and when to follow up. It works, but it’s clunky, and I’ve wanted a proper tool for months. The spec was simple: a login, a form to log a client name, date, and note, and a list view showing upcoming reminders sorted by date.
Not a social network. Not a marketplace. Just a slightly smarter spreadsheet with a login screen. The kind of thing a competent developer could build in an afternoon.
I gave myself one week and five tools: Lovable, Bolt.new, Replit, V0 by Vercel, and Cursor. I used the same plain English description in each one and tracked what happened.
I should mention that I went into this expecting to be impressed. Every demo I’d seen looked like magic, and I am very susceptible to well produced demos. This matters for the results.
Day one: the magic trick
Lovable went first because it was the one I’d seen recommended most often for people who can’t code.
I typed my description into the chat. Four minutes later, there was an app. Login screen, dashboard, a form, a table. It was strikingly attractive, the kind of interface that would look credible in a screenshot. I added a few entries. They appeared in the table. I felt a small, private thrill that I will not apologise for.
Bolt.new was next. Faster than Lovable at the initial build, maybe three minutes. Similar result. A working interface with a login, a form, and a list. The design was slightly less polished but the speed was intoxicating. I started to feel like the articles were right. Maybe anyone could build an app.
Replit took longer because the Agent asked me clarifying questions before it started. What database did I want? Did I need authentication? What should happen when a reminder date passes? I found this slightly annoying at the time. In hindsight, it was the only tool that treated my vague description as vague.
V0 built a gorgeous interface in about two minutes. Beautiful components, clean layout, the sort of thing you’d see in a design portfolio. No database behind it. No way to actually save anything. I typed a client name into the form and pressed submit and nothing happened. More on this later.
Cursor: I opened it, saw a code editor, remembered who I am, and closed it. The entire interaction lasted about ten minutes, most of which was spent looking at a terminal window and feeling the way I feel when someone explains cricket to me.
By the end of day one, I had three apps that appeared to work and one very pretty wireframe. I went to bed feeling like I’d cracked something.
Day three: the thing that everyone warns you about
Two days of testing is where the story changes.
I started asking Lovable for features that weren’t in the original build. Filter the list by date range. Add a “send reminder” button. Sort by client name. Each request was a new message in the chat, and each one produced results quickly. The problem was that fixing one thing kept breaking something else. The filter worked, but the table stopped loading. I asked it to fix the table, and the filter disappeared. The login screen, which had been fine, suddenly started throwing an error I didn’t understand.
This is what people in the vibe coding community call the “fix and break” cycle, and it is the defining experience of using these tools as someone who can’t code. You can see the app. You can almost use it. But “almost” turns out to be the most frustrating word in software.
Bolt.new had the same problem, but with an additional twist: tokens. Every message to the AI costs tokens, and the cost scales with the size of your project. As the codebase grew, each fix attempt burned through more tokens. I ran out of my free tier before the tool was functional. The path from free to useful runs directly through your credit card, and the toll is not clearly marked.
For the record, I did pay for the Bolt Pro plan at £20/month after running out of free tokens. I then spent another hour trying to fix a bug that the AI had introduced while fixing a different bug. This is when I started to question my life choices.
Replit, which had been the slowest to start, was now the furthest along. Because the Agent had asked those annoying questions on day one, the foundation was more solid. When I asked for date filtering, it added date filtering without breaking the login screen. When I asked for sorting, the sorting worked. It wasn’t perfect. The interface was less attractive than Lovable’s. But it functioned, in the way that a tool needs to function when you’re going to open it at 4pm on a Wednesday and trust it with actual client data.
Which leaves V0 and Cursor, and I should talk about both of them because I think there are misunderstandings worth clearing up.
V0 is not an app builder. It’s a component generator. It produces beautiful React interface elements: buttons, forms, tables, cards. These elements look like an app the way a film set looks like a house. From the front, everything is perfect. Behind it, there’s nothing.
If you’re a developer or a designer who needs to prototype a user interface quickly, V0 is probably brilliant. I’ve seen developers praise it and I believe them. But it’s consistently listed alongside Lovable and Bolt and Replit in “vibe coding tool” comparisons, and for someone who doesn’t code that’s misleading. I spent about twenty minutes with it before I understood that I couldn’t build what I needed, and then another ten minutes feeling silly for not realising sooner.
Every comparison of vibe coding tools includes Cursor, and I’m including it too, but only to say this: Cursor is an AI powered code editor. It’s for people who already code. If you are a developer, by all accounts it’s remarkable. Multiple developers I know use it daily and speak about it the way people speak about tools that have actually changed how they work.
But listing it alongside Lovable and Replit in a “build apps without coding” article is like listing a professional kitchen alongside meal kit delivery services. They are solving different problems for different people. I closed it after ten minutes. If you code, ignore everything I’ve just said about Cursor. I am not qualified to review it.
What £55 gets you
By the end of the week, here’s where I’d landed:
Replit produced the only tool I’d actually consider using at work. The interface isn’t going to win any design awards, but the app loads, the data saves, the reminder dates sort correctly, and when I showed it to a colleague, she asked me to share the link. Total cost: about £14 for the starter plan.
Lovable produced the best looking app and the most frustrating experience. Every screenshot I took could go in a pitch deck. The app itself crashed when I tried to filter by date range. Total cost: £20 for the pro plan plus some wasted free credits.
Bolt.new ate tokens like a parking meter eats coins and left me with a project full of code that a developer friend described, after a brief look, as “chaos.” His exact word. He then asked if I’d written any of it myself, and when I said no, he said “obviously.” Total cost: about £20 across free and paid tiers.
V0 taught me the difference between a user interface and an application. Total cost: £0 on the free tier, plus some dignity.
Cursor taught me that I still can’t code. Total cost: £0 and ten minutes.
The gap nobody talks about
Here is the thing that surprised me most, and it’s not about any individual tool.
The marketing for vibe coding says “anyone can build an app.” The more accurate version is “anyone can start an app.” The distance between those two sentences is enormous, and it’s where all the frustration lives.
Every tool I tested was better at producing screenshots than software. The screenshot you’d post on LinkedIn is perfect. The demo you’d show an investor is convincing. The app you’d actually use on a Wednesday afternoon to log a client call? That’s a different thing entirely, and getting from the screenshot to the working tool is where you discover that software development is hard for reasons that have nothing to do with typing code.
The code these tools generate almost works. And “almost,” as I learned this week, is the most expensive word in software. It’s more expensive than “doesn’t work,” because “doesn’t work” tells you to stop. “Almost works” tells you to keep going, keep prompting, keep paying for one more round of tokens, keep watching the AI fix one thing and break two others.
A 2026 analysis of about a thousand Reddit comments about vibe coding found that deployment and debugging are the two most common pain points, even among technical users. That matches my experience exactly. Getting the first version was easy. Getting the first version to actually do what I needed was the part where everything slowed down.
I don’t want to be unfair. These tools are doing something that would have been science fiction three years ago. But the gap between “this is amazing” and “I can use this” is real, and most of the articles I’ve read about vibe coding don’t spend enough time in that gap.
Who this actually works for
The most honest recommendation I can give, based on one week and one simple project, is this: vibe coding tools are brilliant for building something you can show to someone. A prototype. A proof of concept. A “here’s what I’m thinking, can you build the real version” conversation starter.
For that purpose, Lovable is probably the best starting point. It produces the most polished output and the chat interface is the least intimidating. If you need something that works reliably, Replit is the better bet, though it asks more of you upfront.
The workflow that seems to make the most sense in 2026, for someone without a coding background and with an idea, is: use Lovable or Replit to prototype something you can show to a real developer. Then pay the developer. That’s not the “anyone can build an app” story that the marketing tells, but it’s a much better story than what we had two years ago, when the only option was to pay the developer from scratch.
The tools are real. The magic is real, for about the first four minutes. After that, software development reasserts itself, and it turns out that the hard parts were never about typing.
What I expected: To build a working internal tool without knowing how to code.
What actually happened: I built one that sort of works (Replit), two that look like they work (Lovable and Bolt), and learned that “anyone can build an app” means something closer to “anyone can start one.”
What I kept: The Replit project. I’m going to show it to a developer and ask what it would cost to turn it into something my team can actually rely on.
London based, perpetually caffeinated, still trying to leave work by 5:30. I have no sponsors and no one sends me free products. I write because I’m curious about how work actually works, and I have a stubborn belief that someone should try the thing before recommending it. If even one sentence here gave you the feeling of being understood on a bad Tuesday, the whole thing was worth it.
Thanks for being here. 🙏
