Traversal Raises $120M Series B at $1B for AI SRE Agents
AI writes more code, so production breaks more often, and someone has to answer the page at 3am. Traversal announced $120M in new funding at a $1B valuation on Tuesday, co-led by Kleiner Perkins, Redpoint and Khosla, with Sequoia, Battery and Amex Ventures participating. Total raised is $173M. The company came out of stealth only in June 2025.
The product is agents for site reliability engineering, the outer loop that the coding agent boom mostly ignored. The founders trained in causal machine learning at MIT and Berkeley, and that is the bet: finding why a system broke is a causal question across messy, company-specific telemetry, so it is harder than generating a patch. Company-reported traction: ARR up 9x this year, more than 150,000 investigations last quarter, over 4,000 engineers supported. Customers named include Nubank, Harvey, DigitalOcean, PepsiCo and American Express.
Discount what needs discounting. All those numbers are self-reported. The Nubank quote in the release itself says five weeks into the rollout and "it's early". And the most striking number in the release comes with a conflict: on ORCA-bench the best of five frontier agents reached just 25.3% accuracy on medium-difficulty incidents and 10% when incident reports were vague. That benchmark has 1,079 root-cause tasks and was built by Traversal with Cornell Tech and Columbia, so a vendor is grading the competition that it says is not ready.
Even so, the gap it claims is plausible, and the investor thesis has logic. Anthropic and OpenAI have both had to rebuild source control for agent load, and this week's coordination papers say parallel agents still step on each other. More code from more agents means more incidents with no single human who remembers why anything was written.
The thing to watch is whether model labs treat this as their next floor. Coding agents already read logs. If a general agent plus good harness closes that 25.3%, a $1B specialist has a short window. If the moat really is per-customer production context accumulated over time, it does not.
Announcement: https://www.traversal.com/blog/traversal-series-b-120-million-ai-sre Release: https://www.financialcontent.com/article/bizwire-2026-11-3-traversal-raises-120m-with-a-1-billion-valuation-to-keep-code-running-with-a-superintelligent-ai-site-reliability-engineer
← Back to all articles
The product is agents for site reliability engineering, the outer loop that the coding agent boom mostly ignored. The founders trained in causal machine learning at MIT and Berkeley, and that is the bet: finding why a system broke is a causal question across messy, company-specific telemetry, so it is harder than generating a patch. Company-reported traction: ARR up 9x this year, more than 150,000 investigations last quarter, over 4,000 engineers supported. Customers named include Nubank, Harvey, DigitalOcean, PepsiCo and American Express.
Discount what needs discounting. All those numbers are self-reported. The Nubank quote in the release itself says five weeks into the rollout and "it's early". And the most striking number in the release comes with a conflict: on ORCA-bench the best of five frontier agents reached just 25.3% accuracy on medium-difficulty incidents and 10% when incident reports were vague. That benchmark has 1,079 root-cause tasks and was built by Traversal with Cornell Tech and Columbia, so a vendor is grading the competition that it says is not ready.
Even so, the gap it claims is plausible, and the investor thesis has logic. Anthropic and OpenAI have both had to rebuild source control for agent load, and this week's coordination papers say parallel agents still step on each other. More code from more agents means more incidents with no single human who remembers why anything was written.
The thing to watch is whether model labs treat this as their next floor. Coding agents already read logs. If a general agent plus good harness closes that 25.3%, a $1B specialist has a short window. If the moat really is per-customer production context accumulated over time, it does not.
Announcement: https://www.traversal.com/blog/traversal-series-b-120-million-ai-sre Release: https://www.financialcontent.com/article/bizwire-2026-11-3-traversal-raises-120m-with-a-1-billion-valuation-to-keep-code-running-with-a-superintelligent-ai-site-reliability-engineer
Comments