Most AI incident tools still leave a key decision to humans

An AI SRE agent can help troubleshoot an incident. But can you trust it to decide on its own when something needs attention?

In The New Stack‘s October 22 live webinar, Raaz Dwivedi, co-founder and CTO of Traversal and an assistant professor at Cornell Tech, will explore why most AI SRE tools still fall short of that level of autonomy — and what it takes to get there.

Drawing on his team’s work applying causal AI and agentic reasoning to production incidents, he’ll discuss the next steps toward self-healing software.

The three phases of agent autonomy

When we think about agent autonomy, we tend to see it as all-or-nothing: either an agent is autonomous or it isn’t. But for incident response, the reality is more like a three-phase progression.

The first phase is invoked: copilots, chat interfaces, and CLIs that wait for an on-call engineer to open them and ask a question mid-incident. The second is background: scheduled jobs, alert rules, and triggers that fire when conditions are predicted. Both phases depend on a human noticing that something is wrong and deciding to act.

The third phase marks an important shift: proactive agents decide for themselves when an incident needs attention, without waiting to be invoked or scheduled.

What changes once agents act on their own

With many teams and tools stuck in one of the first two phases, Dwivedi will discuss where the shift to full autonomy is already happening, drawing on Traversal Workers already operating at the proactive end of the progression. He will also explore what changes engineering teams will need when the agent in their stack stops waiting to be asked.

The session closes with live Q&A, so you can ask him your questions directly.

Join us on Thursday, October 22, 2026 at 1 p.m. Eastern / 10 a.m. Pacific, to learn where your team stands in the progression and what it would take to move up. Register today.