4/6/2026 Bruce Adams
The Laude Institute, an organization focused on accelerating and funding impactful work in CS and AI, has selected a project let by Illinois CS students for the Slingshot program. Their project, SREGym, is an AI-native platform which provides a live training ground for AI SRE agents with high-fidelity failure drills.
Written by Bruce Adams
Can AI autonomously resolve cloud outages?
A group of students at The Siebel School of Computing and Data Science in The Grainger College of Engineering at the University of Illinois Urbana-Champaign has an answer.
The project develops a benchmarking system for designing, developing, and evaluating AI agents for Site Reliability Engineering (SRE) on high-stakes infrastructure work. It was selected by the Slingshots program from Laude Institute among 14 projects from Stanford, Berkeley, MIT, and CMU. Laude Institute is an organization focused on accelerating and funding impactful work in CS and AI. The Slingshots program supports researchers and scientists with no-strings-attached resources to help them turn research ideas into real-world impacts.
The SREGym project is led by CS students Jackson Clark (CS PhD), Yiming Su (CS PhD), Lily Gniedziejko (CS undergraduate), and Saad Mohammad Rafid Pial (incoming CS MSCS) at the Siebel School of Computing and Data Science in the Grainger College of Engineering at the University of Illinois Urbana-Champaign.
SREGym is an AI-native platform that provides a live training ground for AI SRE agents with high-fidelity failure drills. In SREGym, AI agents are asked to diagnose and mitigate failures with different patterns in different scenarios in a live cloud-native system environment. SREGym provides a comprehensive benchmark suite for evaluating AI-driven SRE agents and for training next-generation agentic AI for SRE. The project is maintained as an open-source project on GitHub with many contributors. It is used by research groups from Microsoft Research, University of Washington, University of Toronto, let alone groups from the University of Illinois. It is also selected as a key benchmark for the System Intelligence initiative from Microsoft Research.
This is a very special group of students who are willing to think wildly and execute in hard ways.
— Tianyin Xu, associate professor, computer science
Watch Jackson Clark's Slingshots presentation at NeurIPS2025's Laude Institute event.
As Clark describes it, “Modern computing systems, no matter for cloud services or for AI infrastructures, are rapidly increasing in their scale, complexity, and dynamics; they inevitably break more often. While AI for coding has taken off in the past year, AI for production systems hasn’t progressed in the same way. Evidence shows that increased use of coding agents is already leading to more incidents in production.”
The project was a spin-off from Clark and Su’s research project on building safe and effective AI agents for system reliability engineering in production.
AI in production is a fundamentally different problem from AI for coding. AI hallucinates. For coding agents – this is fine. You can simply undo its work. However, when a system is live, this is far more difficult.
— Jackson Clark
To evaluate their agentic SRE research, Clark and Su found that there wasn’t a good SRE benchmark for modern agentic AI. Existing benchmarks are overly simplistic and hard to extend; few are open or easy to use.
“We believe that the lack of a high-quality AI SRE benchmark is a fundamental missing piece that blocks or slows down the development of AI-for-production technologies. However, building an effective AI for SRE benchmark is highly challenging – How to create a realistic production environment in an experimental setup? How to simulate realistic faults and expose their failure symptoms to the AI agents? How to evaluate the effectiveness and safety of AI agents? These are the research problems we are trying to solve.” said Su.
Laude Ventures co-founder Andy Konwinski once asked the team what their wild success would look like. “Our answer is to set the standard and unlock a whole new evolution of AI for system reliability engineering,’” said Clark.
Watch the interview by Laude Ventures co-founder Andy Konwinski.
The SREGym project benefits greatly from several undergraduate research programs such as UR2PhD (led by professor Nancy Amato) and UIUC+ Summer Undergraduate Research in SE led by the SE research group (mostly professor Darko Marinov) at the Siebel School of Computing and Data Science, where they found the other two core members, Lily Gniedziejko and Saad Mohammad Rafid Pial.
Gniedziejko details how, as an undergraduate in the UR2PhD program, she joined the team. “I joined xLab (professor Xu’s lab) recommended by professor Darko Marinov. This enabled me to develop a passion for AI for Systems. Additionally, as part of the CS STARS program, I had the opportunity to dive deeper into research and share my work with other students at events like Trick or Research. My favorite part of CS STARS is explaining my passions to other students and helping them find research they can develop a passion for as well. I hope that other students can find an academic home as I have with xLab.”
“Lily is amazing!” Xu observes, “Last summer when Lily joined the group, she hadn’t heard of Kubernetes. A year later, she is leading several initiatives in SREGym with new ideas and creative engineering. She works extremely hard and treats the lab as a home.”
Marinov points to the importance of mentoring when he says,” Illinois students not only produce excellent technical results but also mentor others to produce such results. For example, Jackson and Yiming mentored Lily and Pial, who started research as undergraduate students. Both of them actively serve undergraduate programs in the Siebel school.”“We’re trying to pay back to these programs,” said Su, “we gave talks at student research programs like Undergraduate Research Open House and welcome students to contribute to the project.”
“We also would like to thank professor Brighten Godfrey who has been giving us great advice. The startup seminar taught by him played a pivotal role in encouraging and helping our application.” said Clark. Brighten Godfrey is a professor of Computer Science at Siebel School and research advisor at Laude Institute. He taught the Computing Startup Seminar, which Clark took in Fall 2025.
Grainger Engineering Affiliations
Nancy Amato is professor of computer science, director of the Siebel School of Computing and Data Science. Nancy Amato is Abel Bliss Professor of Engineering.
Brighten Godfrey is an Illinois Grainger Engineering professor of computer science and is affiliated with the Coordinated Science Laboratory.
Darko Marinov is an Illinois Grainger Engineering professor of computer science.
Tianyin Xu is an Illinois Grainger Engineering associate professor of computer science