On-call Agents

Our vision is that AI agents would eventually replace the most disliked part of the software engineer’s job which is to wake up a 2 am in the morning and work on an on-call issue. So this weekend, we built a fully autonomous AI agentic system that handles on-call duties of an engineer. Our agents detect, diagnose, and fix production issues in real time without human intervention.

Running code: In terms of running code, all of the code we wrote is available in a public repository here: https://github.com/hanuytayal/AITinkerers. Also, we will show the live demo of our project to the participants and our demo video is also based on the live run.

Innovation & creativity: Our solution replaces the complete lifecycle of a live incident / production level issue with an agentic system. Our agents go all the way from detecting issues in real-time, analyzing the severity of the issue, diagnosing the problem, and then generating and even executing steps (such as running CLI commands and making changes to your configuration) and then reporting back on what it did in the ticket.

Real world impact: Our solution has a positive impact on developer well being. A heavy on-call burden is the number one reason for engineer attrition and burnout. It cuts engineering toil and helps teams stay focused on building rather than firefighting. Also, in economic terms, our solution can increase the velocity and quality of fixes and reduce Mean Time to Recovery (MTTR) which is an important engineering metric.

Theme alignment: We build AI agents with highly practical real world application. We talked to 20 fellow engineers this weekend and they all complained about on-call and they encouraged us to solve this real world problem.

We worked on a earlier iteration of this idea last week. In an earlier iteration, two of the team members met at a different hackathon and created a related version of this project. In that version, we only used the LLMs to create a diagnosis tool and then suggests which lines of code may have led to the error. We did not build a UI version and we did not build a log analyzer at that time. None of the code was re-used for the AI Tinkerers project. Here is the GitHub repo from the other hackathon: https://github.com/hanuytayal/NoMoreOnCall