🔥 Traversal Raised $120 Million for an AI That Cleans Up After Your Other AIs

Here’s the number that explains the whole deal: $120 million. That’s what a two-year-old startup called Traversal just raised in a Series B, at a $1 billion valuation, to build an AI that does one unglamorous job — figure out why your software broke (The Next Web).

The round, which closed on November 4, was co-led by Kleiner Perkins, Redpoint Ventures, and Khosla Ventures, with Sequoia, Battery Ventures, and Amex Ventures also writing checks. Kleiner Perkins and Sequoia both increased stakes they already held (Goodwin). It takes Traversal’s total funding to $173 million and mints a new unicorn in a corner of the industry nobody puts on a conference T-shirt: site reliability engineering, the 3 a.m. art of answering “what just went down and why.”

The twist is who’s keeping Traversal busy. A lot of the broken code it’s being hired to diagnose was written by other AI.

🧠 Why This Matters

Software teams have spent two years letting AI coding assistants write more and more of their code. That part worked. The part nobody budgeted for is what happens after the code ships and misbehaves in production, where an AI-generated bug doesn’t look like the human kind.

“AI-written software breaks in fundamentally new ways,” says Traversal co-founder and CEO Anish Agarwal.

That’s the pitch, and investors clearly bought it. When a human writes a function, another human can usually reason about how it fails. When a model generates thousands of lines across a sprawling system, the failure modes get stranger, the author isn’t around to ask, and the on-call engineer is left staring at a dashboard at midnight. Traversal’s wager is that the cleanup crew for the AI coding boom is itself a billion-dollar business. The thesis: if AI is going to write the bugs, someone’s going to get paid to find them.

📊 Deep Dive

Traversal doesn’t sell a chatbot that suggests fixes. It sells autonomous agents that investigate. When an alert fires, a “Traversal Worker” starts digging on its own, before a human has opened their laptop.

Under the hood are two pieces the company likes to name-drop. The Production World Model turns a mess of telemetry, code, and deployment history into a map an AI can actually read. The Causal Search Engine then tests thousands of possible causes in parallel instead of chasing one hunch at a time. A fourth component, Traversal MCP, feeds what it learns back to the coding agents upstream, so the same bug theoretically doesn’t ship twice.

The usage curve is the most convincing part of the story:

• Annual recurring revenue grew 9x so far this year (Agarwal declined to share the dollar figure) (The Next Web)
• Monthly investigations jumped from 513 to nearly 73,000 — more than a 100-fold climb in a year (arr.club)
• Customers ran more than 150,000 investigations last quarter alone
• Roughly two-thirds of usage now goes beyond root-cause analysis into alert triage, release-risk checks, and security
• Named customers include Nubank, American Express, PepsiCo, DigitalOcean, and Harvey, with Forbes adding Blackstone to the list

The founding team reads like a causal-inference faculty lounge: Agarwal holds an MIT PhD and a faculty post at Columbia and Cornell Tech, co-founders Raj Agrawal and Raaz Dwivedi bring PhDs from MIT and Berkeley, and Ahmed Lone came out of quant trading at Citadel Securities (The Next Web).

“It’s already becoming part of how our on-call engineers work incidents,” said Nubank CEO David Vélez, whose bank is one of the reference customers.

⚠️ The Catch

Here’s where you should slow down. The same company published a benchmark that quietly undercuts the hype, and it’s worth reading.

Traversal and a group of academic partners built ORCA-bench, a test of 1,079 real root-cause tasks. The best of five frontier AI agents correctly identified the cause of a medium-difficulty incident just 25.3% of the time. Feed those agents vague incident reports — which is what real engineers usually file — and accuracy collapsed to about 10%, while hallucination rates ran as high as 40% (The Next Web).

Sit with that. An industry benchmark, from the vendor itself, says today’s best AI debuggers are wrong three-quarters of the time on moderate problems and hallucinate nearly half the time when the inputs are messy. Traversal’s own system is tuned to beat those baselines, but the numbers are a useful cold shower: this is early, the failure cases are real, and “autonomous incident response” still leans on humans to catch the confident wrong answers. A $1 billion valuation is a bet on the trajectory, not the current hit rate.

🎯 What Happens Next

Traversal says the new money goes toward hiring across research, engineering, and go-to-market, aimed at pushing its agents toward more autonomous production operations — in plain terms, resolving incidents with fewer humans in the loop (Goodwin).

Expect the incumbents to notice. Observability platforms like Datadog and the big cloud vendors already own the dashboards where your alerts live, and “AI SRE” is exactly the layer they’d want to absorb rather than cede to a startup. The real question for 2027 is whether Traversal’s causal approach stays a genuine moat or becomes a checkbox feature inside tools you already pay for. Nine-times revenue growth buys you a head start; it doesn’t buy you the finish line.

🧩 Bigger Picture

Step back and the deal is a tidy snapshot of where the AI boom actually is. The first wave sold you tools to produce more — code, copy, images — faster than ever. Traversal is part of a second wave selling you tools to manage the consequences of all that production: the debugging, the reliability, the security review, the cleanup.

That’s a quieter market than chatbots, but it may be a stickier one. Downtime has always had a price tag, and the more of your stack an AI writes, the more someone needs to understand it when it falls over at 3 a.m. Traversal raised nine figures betting that understanding is the scarce commodity now — and that the companies shipping AI-written code fastest will be the first to pay up for it.

The coding bots got the headlines. The janitor AI just got the bigger check.

---

Sources

• Traversal raises $120M at a $1B valuation to fix what AI code breaks — The Next Web
• Goodwin Advises Traversal on $120 Million Series B — Goodwin Procter
• Traversal grew ARR 9x, raised $120M Series B — arr.club
• Traversal raises $120M at $1B valuation to automate incident response — Dealroom

❓ Quick answers

How much did Traversal raise and at what valuation?

Traversal raised a $120 million Series B at a $1 billion valuation, announced November 4, 2026, bringing total funding to $173 million (The Next Web, Goodwin).

What does Traversal actually do?

It builds autonomous AI agents for site reliability engineering - they investigate production software failures, find root causes, and triage alerts, with customers including Nubank, American Express and PepsiCo.

How accurate are AI debugging agents today?

On Traversal's own ORCA-bench, the best of five frontier agents correctly identified only 25.3% of medium-difficulty incidents, falling to about 10% with vague reports and hallucinating up to 40% of the time.