Conference Presentation, Keynote
Self-Driving Production: AI Wrote Your Code. AI Should Fix It, Too. | Traversal | RAISE Summit 2026
- Traversal, a startup founded by MIT PhD and Columbia University professor, builds an AI site reliability engineer (SRE) focused on solving root cause analysis via causality rather than correlation.
- The company addresses a critical shift in software engineering workflows: while AI tools (e.g., GitHub Copilot, Cursor) have streamlined development, they have inadvertently increased system complexity, forcing engineers to spend more time on production troubleshooting than design.
- Global downtime costs the Fortune 2000 approximately $400 billion annually, with 41% of incidents initially detected by customers, leading to churn and reputational damage.
- Approximately 20% of an engineer's time is currently consumed by maintenance and troubleshooting, a ratio that has worsened exponentially as systems generate petabytes of daily telemetry.
- Current observability vendors (e.g., Datadog, Dynatrace, Splunk) primarily visualize correlations and concurrent failures but cannot explain causation, necessitating large "war rooms" with hundreds of personnel.
- Traversal applies an L0-L5 autonomy framework to production reliability, moving from manual debugging (L0) to full self-healing and prevention (L5), with current industry leaders typically capped at L3 (agentic troubleshooting of single components).
- The core technical challenge involves analyzing petabytes of data (1–10 PB daily) to trace causal links across 10–20 hops of distributed systems, a task impossible for LLMs via raw context window limits.
- Traversal's platform achieves over 80% accuracy in root cause identification within five minutes while processing trillions of logs and spans across customer environments.
- The solution delivers full agentic SRE capabilities, including incident Root Cause Analysis (RCA), alert triage, autonomous system healing, production validation, and "production-aware" code generation via an MCP server.
- Pepsi reduced backlogged alerts by identifying historical alert combinations that could predict incidents, preventing customer-facing outages and saving engineering hours.
- American Express reduced Mean Time To Resolution (MTTR) from 60 minutes to 25 minutes for specific incidents; the system autonomously identified remediation steps (traffic failover, Redis flushing, pod deletion) and involved fewer engineers.
- Cloudways achieved full autonomous healing by remotely removing abusive IPs to resolve incidents without human intervention.
- Customers utilize Traversal as a unified "production memory," handling thousands of daily queries for change validation and observability improvement beyond incident response.
- To achieve enterprise-scale AI deployment, Traversal emphasizes answering five critical questions: (1) accessing complete production data without gaps; (2) reasoning at scale without prohibitive LLM/egress costs; (3) real-time entity and relationship mapping for AI agents; (4) autonomous system self-improvement without manual fine-tuning; and (5) ability to identify root causes far from initial symptoms.
- The company's proprietary assets include a "production world model" for data representation and a "causal search engine" agentic harness designed to navigate complex system dependencies.