I’ve been experimenting with AI coding agents recently, and the speed at which I can move between different layers of the stack is mind-blowing. Today, I sat down with an AI agent (Google Antigravity in my case but can be any like Cursor or Claude Code) to tackle a laundry list of tasks that would normally take me half a day. We finished them in under an hour.
Here’s a breakdown of the “flow” and what we accomplished in a single session.
1. Fixing Legacy Code (Python/AWS Lambda)
I started with a bug in the email forwarder for cloudyeti.io, my personal cloud project. It was sending emails as attachments instead of inline text.
- The Task: Modify a Python Lambda function to parse MIME parts and reconstruct the email body.
- The AI’s Role: I didn’t have to look up the
email.messagelibrary docs or remember how to traverse multipart payloads. The agent just wrote the code, zipped it, and updated the Lambda function.
2. Debugging CI/CD Pipelines (GitHub Actions)
Next, I switched context completely to blog.bipratech.com, a site for my dad’s company. The deployment workflow was stuck.
- The Fix: The agent identified we were using an old macOS runner and switched it to
ubuntu-latest. It also fixed a subtle S3 bucket naming error and removed a deprecated ACL flag that was causing the build to fail. - The Result: A green build and a successful deployment to CloudFront.
3. Frontend & Design (Saurav.io)
Then came the creative part. I wanted a new, sleek portfolio for saurav.io.
- The Build: I gave the agent a rough idea (“minimalist, dark mode, sleek”). It generated a single-page HTML/CSS site with a premium feel, “Inter” typography, and a responsive design.
- The Deployment: We synced it to S3 and invalidated the CloudFront cache.
- The “Oops” Moment: I realized the site wasn’t updating. The agent investigated Route53, found that
saurav.iowas actually pointing to a different S3 bucket (sauravsharma.net), and corrected the deployment target instantly.
4. Rapid Iteration
Finally, I wanted to tweak the content—remove the fluff, keep it minimal (“Cloud Technologist • AI Builder • Indie Musician”), and add a profile picture.
- The Speed: These changes happened in seconds. No context switching, no “where is that file again?”, no “what’s the CSS for a circular image?”. Just intent -> execution.
The Takeaway
Tools like Google Antigravity, Cursor, and Claude Code aren’t just “autofill” anymore. They are context-aware collaborators. They allow me to be a “Cloud Engineer” one minute, a “DevOps Engineer” the next, and a “Frontend Designer” right after.
The bottleneck is no longer writing the code; it’s having the clarity of what you want to build. When you have that, the AI handles the rest.