Ciroos is building AI SREs that can actually fix things

Ronak Desai, CEO & founder of Ciroos, has built five startups. His last one scaled to $6 billion inside Cisco. He ran AppDynamics as SVP and was part of the Splunk due diligence. He knows observability. And he knows the fundamental problem hasn't been solved: mean time to repair is stubbornly high.

We've gone from grepping logs to monitoring tools to full observability platforms. Yet engineers still get woken up at 3 AM, still scramble to look at 10,000 dashboards, still deal with incredibly stressful on-call rotations. Ciroos is betting that AI agents can finally close that gap.

The multi-agent architecture

Ciroos built a multi-agent system with specialized agents for different domains: network, security, cloud environment, application performance, Kubernetes. Instead of building one massive data platform and hoping for anomaly detection, they bring intelligence to where the data already lives.

The goal: before anyone gets on a call, Ciroos has figured out what happened, identified the root cause, collected evidence, and provided remediation steps. If you give them access to your CI/CD pipeline or Git repo, they'll tell you exactly which change broke things and generate a PR to fix it.

Human in the Loop (for now)

Ronak believes 2025 is the year of agents, but no serious enterprise is ready to let agents run completely free. Human in the loop is critical. Ciroos provides recommendations and evidence, but humans make the final call on taking action.

That said, he sees a clear trajectory. Start with read-only access, build trust, then move toward autopilot mode. He compares it to Tesla's evolution—from basic lane following in 2015 to where we are now with Waymo driving around cities. The path from augmentation to autonomy won't take ten years in software.

Nine months from zero to enterprise production

Ciroos started in February 2025. By re:Invent, they're deployed in enterprise production environments. This timeline would have been impossible even five years ago.

Three things made it possible. First, AI tools like Cursor and Claude Code let the team build faster—Ronak estimates they're 50% ahead of where he expected to be. Second, he's contributing code again after years away because the tooling makes it accessible. Third, the SaaS ecosystem has matured. Authentication, authorization, all the enterprise requirements—partners like WorkOS make these table stakes rather than multi-year projects.

The surprising enterprise pull

What surprised Ronak most is the pull from large enterprises. Companies with 30,000 employees are coming to them saying they need AI deployed to bridge the gap between what needs to be done and the staff available to do it.

This isn't just cost optimization. The quality of AI-generated investigations and recommendations exceeds what even distinguished engineers can produce. If you're a financial institution that can't hire that level of talent, AI agents become the great equalizer.

The flip side is happening at small companies too. Startups with 50-60 people who have no dedicated SREs—just devs doing ops—can now get started with best practices for reliability, redundancy, and autoscaling without hiring specialists. Ciroos is augmenting at the top and enabling at the bottom.

This interview was conducted at AWS re:Invent 2025.