MIT AI Expert: AI Is Writing Code Nobody Understands | Ep 50

What happens to observability when the person staring at the dashboard can no longer understand the code an AI just wrote? For twenty years, monitoring tools have been built on one assumption, that a human is watching. As AI writes more of the code and the telemetry, that assumption may be quietly collapsing.

In this episode of Inside the Silicon Mind, we sit down with Anish Agarwal, CEO and co-founder of Traversal, to dismantle the model of observability that the industry has relied on for two decades. Anish is a professor at Columbia University with a PhD from MIT, where he and his co-founders researched causal AI and reinforcement learning, teaching systems to reason about cause and effect rather than just correlation. He left a career he never intended to leave after concluding that the most interesting research now happens inside companies with research in their DNA. Today, Traversal is an agentic AI SRE platform that troubleshoots and self-heals complex systems for enterprises producing roughly a petabyte of telemetry a day.

You’ll hear why dashboards are really just enshrined queries, snapshots of how systems broke in the past that struggle to predict how they will break in the future, and why root cause analysis resembles finding a needle in a haystack full of fake needles when one broken service spreads like an epidemic. Anish explains why he believes the sixty-person war room should not and will not exist, with agents handling most incidents and pulling in one to five humans only when confidence is low. He also makes the case that this is fundamentally an AI problem rather than an observability problem, which is why he thinks fragmented, ingest-priced incumbents will struggle to respond.

Who this episode is for:

• SREs, DevOps, and platform engineers wondering how agentic AI will reshape on-call, incident response, and their careers
• CTOs and VPs of Engineering responsible for reliability at enterprise scale
• Founders building AI-native infrastructure products in markets dominated by incumbents
• Investors evaluating the observability market and where value shifts when AI consumes the telemetry
• Researchers and academics weighing the move from universities into startups

What you’ll learn:

• Why dashboards are enshrined queries that explain the past and may fail to predict novel failures
• How causal AI separates root causes from symptoms when every alert looks like a fake needle
• Why Anish believes the war room is ending, replaced by agents that escalate to just one to five humans
• How SREs could shift from firefighting to architecting resilient systems, even as AI-written code pushes engineers toward less creative review work
• Why incumbents may struggle to win: talent density, ground-up stack redesign, and pricing models tied to data ingestion