GPT-5 Hands-On: Vibe Coding, AI Automation, and the Rise of VibeOps |

This is a big one. This article explores GPT-5's new coding capabilities, the future of AI automations, and a new concept called VibeOps.

First, the big hype: GPT-5 is here, it's free, and it's available right now in Cursor. For its launch week, you can get over a quarter of a million tokens in free credits. If you can't wait, you can always use GPT-5 fast. It's the new way to vibe code.

Building a Real-Time GPT-5 Hype Dashboard

To start, I prompted the AI to build a dashboard to track real-time mentions of OpenAI's GPT-5, complete with beautiful rainbow visuals and dopamine-inducing explosions. The goal was to display real information about what's being said right now about GPT-5.

To make this work, the plan was to hook it into the real-time Grok API, which has a live search feature. I provided the API documentation, specifically the guide on Live Search, to ensure it got the context right. This was my first time using GPT-5 to vibe code.

Here was the initial prompt:

I'd like you to build me a dashboard that tracks, in real time how much OpenAI's chat GPT-5 is being spoken about right now, and it demonstrates it in beautiful rainbow visuals with dopamine inducing explosions and pops and panels, and also gives real information on what is being said right now about GPT-5. Use this API documentation.

The AI began planning a real-time aggregator. Without any framework or specific context provided, it decided to scaffold a React front end and an Express backend. After about 30 seconds of planning, it started executing commands, working on the dopamine explosions, inserting API key variables, and creating Readme files. It felt like having a real developer in my back pocket.

After providing my xAI API key, the result was a dashboard where the GPT-5 hype was off the charts with rainbows, confetti, and positive, sensible keywords. Real-time items were coming in from the xAI API, showing a significant amount of positivity. This entire dashboard was created with just a single prompt.

Creating a 3D Pinball Game with AI in Just 5 Minutes

Next, I decided to continue the vibe test by asking the AI to create a game in three.js.

The prompt was:

I'd like you to make me a game in three.js called Pinball Planets. It will be a pinball machine, but not as we know it. Lots of colors, lots of space and galactic themes. And lots of beautiful planets that I can bounce my ball between.

The AI immediately started planning and setting up the project. In no time, it was ready and provided a URL. The result was something with flippers, but not a traditional pinball machine, and it included 3D physics.

To enhance it, I added another instruction:

Make it so that when I tilt and pan around, that the ball moves and falls in the correct direction, following physics.

After about 90 seconds, the AI implemented live tilt controls and dynamic gravity. The ball now reacted immediately to camera rotation, tilting and rolling around based on the orientation of the view, eventually disappearing off the edge into the vastness of space.

Testing GPT-5 Inside Lovable

Next up, I tested GPT-5 inside Lovable, an AI vibe coding app. After enabling the new model, I created a new database called Health Stats and connected it to my app.

The prompt was made more complex by incorporating a Python library:

I'd like you to build an app that visualizes my most recent Garmin stats, and allows me to chat to them using AI. This should be implemented using the Python Garmin Connect library and my backend database for storage and edge functions. To use the AI functionality. This is the repo you need: https://github.com/cyberjunky/python-garminconnect

GPT-5 kicked into action, writing everything needed for Supabase. While it ran slower inside Lovable, it created significantly more files than the default agent. It planned to build several edge functions to ingest Garmin logs and chat with them using AI, correctly referencing the provided repository.

After adding the necessary API keys, two tables were created in Supabase for daily and regular activities, along with edge functions for Garmin AI chat and data ingestion. The app successfully pulled in my Garmin stats, and a synchronize button was added by the integration.

But could I chat with my stats? I asked:

Tell me my most active day over the last two weeks, and suggest a workout that I could do today to compound the effects of my brilliant exercise over the last two weeks.

The AI replied:

I can see that your most active day over the last two weeks was August 2nd, 2025. I should consider a 30 to 45 minute walk or jog combined with bodyweight exercises, and this will help maintain my cardiovascular fitness. Aim for 30 minutes of activity to keep building on your momentum.

While the result was a basic but functional app, the experience showed the potential of using GPT-5 for such tasks.

The Future of AI Automation: Builder vs. Delegator

The world of AI automation saw a big shake-up this week with pricing plan changes for n8n, a powerful tool for technically minded people. For self-hosters and those with simple automations, the changes are positive, pushing users to the cloud plan and removing execution limits for complex workflows.

However, for high-end power users, a new business pricing plan introduces what some have called a "success tax." If you're a high-end business with numerous AI-powered automations, you'll pay for each execution over a certain limit on your own hardware. This raises questions about the long-term sustainability of the unlimited free model.

But while many argue over the cost of executions versus tasks (n8n vs. Zapier), the entire architecture of AI automation is fundamentally changing.

n8n is doubling down on its architecture for tinkerers and builders, allowing for complex workflows with over 15+ steps and total control. It's the perfect workshop for those who enjoy the craft of building their own engine.
Zapier is playing a different game. They are building a new layer of abstraction with Agents, Chatbots, and MCP, all with generous free plans. They are building the full self-driving car; you just punch in your destination. In the future, you won't think in steps (Trigger, Filter, API call). You'll just tell the Zapier agent, "When a new customer pays on Stripe, onboard them and send them a welcome kit."

This isn't about which tool is better; it's about two different visions for the future. * n8n's vision: Empower the builder who enjoys driving a manual transmission. * Zapier's vision: Empower the delegator who wants to punch in a destination and let the AI take them there.

The real choice becomes clear: Do you want a better tool to build with AI, or do you want a tool that is AI, ready to take your instructions?

Introducing VibeOps: Your Own AI-Powered Server

The idea of being a delegator is expanding to cloud infrastructure. A fascinating new concept, termed VibeOps, emerged this week. The idea is simple: start a VPS, SSH into it, and use AI to configure and manage it. It's basic engineering knowledge to be able to operate your own server, and now it's easier than ever.

Here’s a walkthrough of how to set up your own $5/month server for vibe coding using the VibeOps approach.

Step 1: Get a VPS You can get a cheap Virtual Private Server from providers like Linode (now Akamai), Digital Ocean, or Hertzner for about $5 a month. The process involves launching a new virtual machine.

Step 2: Configure Your Server Setting up is straightforward: 1. Select a region close to you. 2. Choose an operating system. Debian 12 is a solid choice. 3. Select a plan. For Vibe Ops, the cheapest shared CPU plan (e.g., Nanode 1GB) is perfect. 4. Give it a label (e.g., "VibeOps"), a strong password, and optionally add your SSH keys for passwordless login. 5. Click Create.

Your server will be provisioned, and you'll get an IP address to access it.

Step 3: SSH Into Your Server Use a tool like Terminus (available on all major platforms) to connect. Simply copy the SSH command provided by your VPS host and paste it into your terminal client.

ssh root@YOUR_IP_ADDRESS

Once logged in, you're ready for the fun part.

Step 4: Use AI to Set Up the Environment First, update your packages: bash apt update Next, install npm, which is needed for many development tools. bash apt install npm -y Instead of a traditional code editor, we'll install the Cursor CLI to run GPT-5 from the command line. bash npm install -g cursor-cli Follow the on-screen steps to log in. Once done, you'll have the Cursor Agent running on your server, with GPT-5 as the default model.

Now, you can treat it like a hobby server and ask the AI to secure it.

Secure the server with basics like firewall and Fail2ban.

The agent will run the necessary commands. You can even tell it to auto-run commands in the future. Next, ask it to install your development stack.

Install Docker and Docker Compose. Install Node.js, the latest version. Install Python.

The agent abstracts away all the complex commands, installing everything for you.

Step 5: Vibe Code Your First App Now, let's create an app.

Create a new directory called myapp and then CD into it. Create a simple express.js Hello World app. But being me with lots of rainbow animations and explosions, just to make it fun.

The AI will create the app structure, write the code for the packages, and build the application right on your server. To deploy it, you can ask it to containerize the app.

Make a Dockerfile for the app and create a Docker compose and run it with the flag on port 3000.

The agent will write the Docker and Docker Compose files and spin up the container. You can then access your app by navigating to http://YOUR_IP_ADDRESS:3000. The result is a fun, interactive "Hello World" page with rainbow animations and an explosive button.

Note: It's never secure to expose your server's IP address publicly. For a real application, you would use VibeOps to set up a reverse proxy, a domain name, and put it behind a service like Cloudflare for protection.

Testing OpenAI's Free GPT-OSS Model

Finally, let's try out the free and open-source model, GPT-OSS, released by OpenAI. Using Ollama Desktop on a Mac Studio with an M1 chip, I tested the 20-billion parameter model. It was lightning-fast, providing a comprehensive reply to "What is the meaning of life?" almost instantly.

For the next prompt, I used the same command from the VibeOps experiment:

Create a simple express.js Hello World app with lots of rainbow animations and explosions.

GPT-OSS came up with a solid suggestion, writing out the project structure and all the files needed to make the app.