From DevOps ‘Heart Attacks’ to AI-Powered Diagnostics With Traversal’s AI Agents

Overview

This episode explores how Traversal, co-founded by Anish Agarwal and Raj Agrawal, is revolutionizing DevOps and site reliability engineering (SRE) with AI agents. These agents perform root cause analysis (RCA) in minutes, addressing the growing complexity of debugging AI-generated code. Drawing on academic research in causal inference, Traversal automates troubleshooting workflows, reducing the chaos of traditional incident management and enabling engineers to focus on creative, long-term infrastructure planning.

Notable Quotes

- If Traversal does its job right, the constant pain of death by a thousand cuts should be something AI takes care of, leaving DevOps to focus on the creative, fun parts. – Anish Agarwal, on the transformative potential of AI in SRE.

- The way we write logs will fundamentally change—they’ll no longer be meant for humans to scroll through but for AI systems to consume. – Anish Agarwal, on the future of observability.

- We’re at L4 for root cause analysis when the data is there, but L2 when it’s not. The goal is to reach L5 with system-wide fixes. – Raj Agrawal, on Traversal’s progress in automating RCA.

🛠️ The Current State of DevOps and SRE

- Anish Agarwal likens the current state of DevOps to having a heart attack twice a week, with engineers constantly firefighting high-severity incidents and chronic system issues.

- Observability tools today focus on data storage and visualization but leave the complex troubleshooting workflows entirely manual.

- Traversal aims to automate these workflows, allowing engineers to shift from reactive problem-solving to proactive infrastructure planning.

🤖 AI Agents for Root Cause Analysis

- Traversal’s agents use large language models (LLMs) to orchestrate tools like anomaly detection and data processing, enabling them to traverse complex dependency maps.

- The agents mimic human troubleshooting but with systematic, data-driven reasoning, avoiding gaps caused by missing system knowledge.

- In large enterprises, where fragmented teams lack holistic context, Traversal’s agents excel by processing vast amounts of data inaccessible to any single human.

📊 Challenges and Innovations in Observability

- Observability remains fragmented across tools like Datadog, Splunk, and Grafana, with no incentive for interoperability. Traversal’s agnostic approach to data storage offers a competitive edge.

- The company’s architecture relies on inference-time computation, allowing agents to adapt dynamically without hardcoding workflows.

- Traversal’s offline phase builds a rich dependency map using LLMs, statistics, and causal inference techniques, while the online phase applies this map to live incidents.

🌟 The Future of SRE and AI-Driven Engineering

- AI systems will redefine SRE roles, requiring fluency in both traditional reliability principles and AI-specific failure modes.

- Logs will evolve to be AI-readable, embedding richer context for automated analysis.

- As AI-generated code becomes the norm, tools like Traversal will be essential for debugging systems where engineers lack firsthand knowledge of the codebase.

- The ultimate vision includes vibes coding for mission-critical systems, where functionality is validated through robust testing rather than human-written code.

🚀 Building an AI-Native Company

- Traversal’s team is 90% engineers, blending PhD-level machine learning expertise with traditional software engineering.

- The company prioritizes adaptability, making six-month bets on AI advancements to stay ahead of the curve.

- Anish Agarwal emphasizes that success in AI engineering is more about an experimental mindset than formal credentials.