Anish Agarwal, Traversal: Production Software Keeps Breaking, and It Will Only Get Worse

Market Trend: AI-generated code now constitutes approximately 30% of all code written, with Anish CEO predicting this will rise to 80% by next year.

Emerging Problem: As AI assumes more development responsibilities, the software engineering workflow is shifting from building to troubleshooting, potentially increasing the proportion of time engineers spend on maintenance rather than creative design.

Root Cause of Complexity: Troubleshooting is becoming more difficult because AI-written code reduces human engineers' contextual understanding of the codebase while simultaneously increasing system complexity due to higher code volume.

Current Pain Points:

• Enterprises currently spend approximately 20% of their cloud budget on observability tools (e.g., Datadog, Dynatrace, Prometheus).
• Incident response often involves "dashboard dumpster diving," where engineers sift through massive telemetry data to distinguish symptoms from root causes.
• Major incidents at large enterprises frequently require 10 to 50 engineers in "war rooms" and can take days to resolve.
• Traditional AIOps and simple React agents fail due to dynamic environments, false positives, and the inability of current models to process petabyte-scale telemetry data.

Traversal's Technical Solution: The company utilizes a combination of three distinct technological approaches:

• Causal Machine Learning: Filters massive data statistically to identify cause-and-effect relationships rather than simple correlations.
• Reasoning Models: Leverages advanced LLMs (e.g., O3, Sonnet) with increased inference-time compute to perform semantic analysis on log messages and metadata.
• Agent Swarms: Deploys thousands of parallel agents working in an "agentic MapReduce" fashion to test hypotheses simultaneously and converge on solutions efficiently.

Future Roadmap: While currently focused on root cause analysis and remediation, the long-term goal is to automate the full software reliability lifecycle, including alerting, preventative measures, and post-mortems.

Case Study: DigitalOcean:

• Reduced time-to-resolution by 38% across their entire engineering organization over a six-month period.
• Achieved an 82% incident scoping rate with 84% accuracy in root cause identification.

Case Study: Financial Enterprise:

• Successfully identified root causes in under five minutes for an infrastructure processing over 1 trillion logs and 100 billion metrics.
• Integrated fragmented data sources including OpenSearch, Dynatrace, ServiceNow, and GitHub.

Target Customer Profile: The solution is optimized for large enterprises where data is mature and well-instrumented but highly fragmented across multiple teams and systems.

Broader Application: The underlying "needle in a haystack" workflow has potential applications beyond observability, extending to security, networking, product analytics, and business operations.

Company Founders & Team:

• Founded by three PhD alumni from MIT and a fourth co-founder from Citadel Securities.
• The 7-person team includes seven members from Citadel, engineers from DevTool companies (Datadog, ServiceNow, Cockroach Labs), AI product engineers from Meta/Perplexity/Glean, and researchers from top universities.
• The company is headquartered in New York City and is actively hiring.