Abusing AI Pull Request Bots For Fun and Profit

This doesn’t scale — but who cares

This setup isn’t for enterprise—but we don’t need it to be. Most companies wouldn’t enable multiple SaaS tools that all do the same thing for budget reasons or be in a position to share their code base with them. As a lean startup, we’re uniquely positioned to leverage generous AI tool offers, largely because we’re open-source (code is public), have access to YC perks and discounts, and intentionally keep our engineering team small. This is how our tiny team maintains thousands of integrations without breaking the bank. Scaling can wait; speed and efficiency can't. We’re extremely grateful for the generosity of the services we’ve used and their open-source (OSS), YC Startup, and free tiers. 

Our product development process

At OpenInt, we live and die by continuous delivery. With just four engineers—each a full-stack contributor leaning into their strengths in design, frontend, or platform—we rely heavily on feature branches. Every feature or change undergoes CI tests, automated and manual reviews, and a quick visual validation for UI changes. Quality isn’t siloed: it's owned collectively, ensuring high standards without slowing us down.

Supercharging this process

The results are Simply insane

We simply can’t see ourselves going back. Some of these tools make claim such as 3X more bugs caught, 4X faster merges, etc. I don’t have a sufficiently large sample size in our engineering organization to verify these statistics, but here’s what I can say. I have NEVER felt more empowered as an engineer to move quickly. I FEEL less stressed when making changes to our platform as I know we have a solid process watching our backs. 

Watch it in action

Here’s our last 3 days of PRs (team of four)

Here’s our last week of activity 

Here are some crazy nuggets caught by the bots

Notably, this involved a sensitive refactoring of our routing implementation. The kind of work where it's easy to make mistakes due to string mismatches that aren’t caught by the compiler. 

Prophet caught wrong navigation by comparing string hrefs to nextjs directory paths

Recurse enforced our FE style guides taking into account SSR, client side and stream based implementations

Ellipsis Summarized our PR and suggested two improvements.

UI related services like Stably and Chromatic deserve their own walkthrough, possibly with a video so i'll leave this for a future post.

Summary: We are never going back

We’re all-in on AI-enhanced development—and we won’t go back. Beyond impressive metrics and faster releases, integrating these bots has fundamentally changed our approach. It’s not just about speed or fewer bugs; it’s about confidence and creativity. Knowing a meticulous AI-powered safety net has our backs, we ship faster, experiment more boldly, and genuinely enjoy the process.

Want to check out the services mentioned? 

• Recurse: http://recurse.ml/
• Ellipsis: https://www.ellipsis.dev/
• Cursor: https://www.cursor.com/
• Greptile: https://www.greptile.com/
• Prophet: https://try-prophet.com/
• Chromatic: https://www.chromatic.com/
• Stably: https://www.stably.ai/
• Merge: https://www.mrge.io/

They’re ALL awesome. Thank you for reading!