Our initial concept
When I stepped down as CEO from Reloadly in March 2023, I thought it was finally time for a break from the startup grind. Little did I know, our former Head of Engineering, Arun Patra, had the same idea and took his break a month later. We found ourselves chatting on WhatsApp about Reloadly, reminiscing and discussing ongoing engineering hurdles. A recurring theme was the frustration over system-wide issues and bugs, often going unnoticed until a customer had already churned.
Over a few months of these conversations, Arun and I decided to tackle this problem head-on. We envisioned an engineering analytics tool that would act as a co-pilot for engineering teams, helping them make better decisions. Excited by the idea, we jumped right in and started building the platform. Fast forward six months, and we were thrilled to be accepted into Techstars in March 2024.
The Pivot (the feeling of being punched in the face)
However, within the first week at Techstars, it hit us: engineering analytics idea wasn’t as hot as we thought, in fact it was more of a vitamin, not a painkiller. We discovered this after talking to many users, co founders and our ICPS – which were CTOs. We further concluded that we should aim to solve the real, painful problem developers were facing—automating bug fixing. From our experience, bugs were a massive headache, hindering product velocity, killing team morale, and driving users away. We realized there wasn’t a good solution out there that could automate the entire process of error resolution.
So, we decided to pivot. Our analytics platform would be the free carrot, offering insights on DORA and code quality metrics. But our main focus would be automating code reviews and bug fixes in production code. Products like Github Co-Pilot, Codium and Codacy are mostly acting like a code assistant, but actually fixing buggy code in production in an automated way before customers can complain is something in another level of sophistication, and that’s the path we decided to take. Most developers hate the code review process, and often is the case why they speed through it and bugs creep into production. Here’s a great resource on the art of pivoting by YC.
The Challenge
We set a hard deadline: April 14, 2024. We had four weeks to prove our hypothesis and build our MVP. The team worked around the clock, and we delivered our proof of concept on time. Our POC included two key products:
- Automated code review
- Automated bug fixes in production code
The Approach
We decided to focus on one product at a time, aiming to deliver something our customers would rave about. Our first official launch at the end of August will be our AI code review for GitHub pull requests. Yes, there are a few code review companies out there but from our analysis non of them provide a high quality solutions that customers can truly depend on. We stripped away all the clutter and focused on delivering the best code review product possible. For eight weeks, we honed our AI tech stack to produce top-notch, human-like results from code reviews.
Key Features of Astronuts AI Code Review
- PR Summaries: Well-written technical summaries highlighting PR diffs and errors.
- Static Code Analysis: Every pull request includes 12 static code analysis metrics configurable from the customer dashboard.
- Line by Line Comments: We process the entire set of diffs in the pull request, providing 98% accurate error spotting.
- Auto Fix: Astronuts suggests high-quality fixes for bugs and errors detected, with a one-click commit option.
- Labeling: Pull requests are ranked (low, medium, high) and labeled as APPROVED or BREAKING CHANGE based on custom algorithms.
These features ensure consistent, high-quality suggestions in real-world scenarios, focusing on security, scalability, performance, and code understanding. With our finely tuned AI models, we’ve reduced false positives to less than 3%, achieving 97% accuracy. This makes Astronuts the best in the market for AI-driven error resolution. By focusing on delivery one high quality product at a time, we were able to realize that we needed improve our AI stack to obtain the highest quality results for our customers. Developers are still weary about AI when it comes to dealing with code, therefore we wanted to ensure that our code review product returned the high quality resolutions for our users.
What’s Next?
We have exciting features in the pipeline for automated code reviews:
- Custom-built rules for pull request standards
- Chat assistant for pull requests
- IDE plugins for VSCode and IntelliJ IDEA
- Pull request templates
- Usage-based billing—no monthly subscriptions, just pay for what you use
Conclusion
Astronuts is on its way to becoming an end-to-end error resolution platform, automatically fixing bugs throughout the software development lifecycle. We’re releasing our products quarterly, so make sure to sign up and stay tuned. Follow us on X and send us a shout out if you find our product useful, we’d love to get your feedback good or bad that is.
By focusing on what developers truly need and iterating quickly based on user feedback, we’ve managed to pivot swiftly and effectively within just four weeks. It’s been an intense journey, but we’re excited to share the results with the community and help developers tackle their biggest pain points with our automated debugging dev tool.