‍

{{review-toc}}

‍

Over the past year, vibe coding went from meme to workflow: describe what you want, let an AI agent scaffold the app, and steer the result by conversation rather than combing through files. The term itself was popularized by Andrej Karpathy—“forget that the code even exists”—and later sharpened by Simon Willison, who distinguishes vibe coding from vibe engineering, the type of AI‑assisted programming where seasoned developers accelerate their work by using LLMs.

‍

In this article I've partnered up with Emergent, an app builder which makes it very easy to ship a full stack app, and sits at the vibe coding end of the spectrum. To see whether Emergent can ship something real (not just a pretty mockup), I asked it to build a simple but fully functional digital swear jar—the type of jar used discourage people from using profanity. However, I wanted a version that can be used to keep track of (and make you pay up for) for any "bad habits" that you want to avoid. I wanted functionality like user sign in, possibility to create more than one jar, an animation for dropping coins, playful messages and, last but not least, a live leaderboard where you can see what other users are tracking and how much damaged they've done. I didn't want to implement any kind of payment, but rather purely a tracking app where users hold themselves responsible.

‍

You can see the app I made live at jar.whatplugin.ai. To create it I needed Emergent to build all the essentials of an app—auth, a real database, and deployment.

‍

‍

Building a full-stack mini app with Emergent

I started by describing the app I wanted in plain English. Emergent didn’t rush into code; it paused to ask clarifying questions first, which I appreciated. The questions it asks you forces you to plan out and visualise in your head what you want to build (a skill that pays off with all AI coding). A few minutes later it came up with a first version, which looked alright, but needed a way to actually save the label and progress of my jar so that I could continue with the same jar the next time I used the app. I explained what I wanted and it created a database with a working MongoDB setup in a matter of minutes. I then asked for sign‑in and—after a detour I’ll explain later—I had authentication working. The agent nudged me to make a GitHub repo and commit often, which is really the best insurance you can have when coding with AI goes sideways (which it always does at some point).

‍

‍

Setting up user authentication

My biggest frustration with Emergent was self‑inflicted: Coming from CLI-based tools (like Claude Code and Codex) where there is no "built-in" user authentication, I tried to wire a Google sign‑in the way I would from a CLI tool. I spent hours troubleshooting with the agent; it tried, but never got flow really working. Emergent's customer support was responsive (answered in a couple of hours, plus gave me an offer to jump on a call). But the real fix was simple: ask for Emergent's own Auth. The agent quickly wired up a login screen, user model and saved data to our database. Lesson learned: when working with vibe coding tools, if they have a native integration for something: use it. In Emergent's case, you just need to tell the agent to "add authentication" or even simpler, "add user log-in."

‍

‍

What I didn't like

I had two more modest annoyances. The agent “goes to sleep” between sessions, so each time I wanted to resume my coding session (e.g. a day later) it takes a couple of minutes to "wake up". Also, while the agent is working, you can’t keep typing to it. This made me keep a separate notes doc with my notes that I would give it for the next prompt.

Another thing I noticed is that results where much better when I hit the agent with small and specific requests, rather than big prompts. When trying prompts like “add these six features...”) the agent often implemented only half of what I asked for; meanwhile, it did much better for contained and specific prompts like “add a leaderboard of the top ten jars by total coins."

‍

Saving your process and deploying your app

Making your app available to the public (ie deploying it) is easy with Emergent's native solution. It builds and deploys in literally one click, makes it easy to add secrets (API keys, etc) and use your own custom domain if you want. However, their native deployment solution costs about 50 credits per month whereas some platforms offer free deploys for light usage. Also, the GitHub integration that I consider essential for coding with AI sits behind their paid tier—totally fair, but worth noting if you’re building something beyond a simple test, and were hoping to stay entirely on free.

‍

Choosing between models and working with MCPs

Emergent gives you the possibility to choose which tools you want to use for your app. For example, you can choose between the latest LLM models (GPT‑5, Claude Sonnet 4.5, and more). Paid plans also gives access to the deepest reasoning modes or things like bigger context windows. If you prefer other backends than their native MongoDB integration, you can also connect to any external services via MCP. So it's totally possible to set your app up with something like Supabase instead, if you prefer. Again, the path of least resistance is to use Emergent’s defaults and keep building.

‍

‍

What you get for free, and what you'll need to pay for

Let's talk about pricing. Emergent's free tier is fine for dabbling, e.g. you vibe code a couple of times a week. As long as you don't need to save your progress to GitHub or share your app with the world, it works. The Standard plan (around $20/month) makes sense if you want to build with it several times per week and also want your end result to be a shareable app; the Pro plan (around $200/month) is for heavy use and large contexts. For an MVP you actually want to show people, plan on paying at least for Standard.

‍

‍

Alternatives to Emergent

If you want really simple app building like Emergent offers, a close alternative is Lovable. If you want more control, with the ability to see and tweak your code, Bolt or Replit are the next rung of complexity. And if you’re more comfortable with code, the CLI tools (Cursor, Claude Code, Codex) gives you the most flexibility and the best unit economics—but with the steepest learning curve. For a full list of the top vibe coding tools, also check out my one-page guide.

‍

Bottom line (also, try I made!)

Emergent is a straightforward to path to ship a small, real app with auth + database + deploy while using nothing but natural language (zero coding). It’s not the tool I’d use for deeply custom or complex systems, but for getting from idea to protoype, it hits a nice sweet spot.

‍

The app I made, the Swear Jar, is live at jar.whatplugin.ai. Try it for making a jar or two, and check out the leaderboard.

‍

‍