Traversal

Behind every new Traverser is a lot of work from our Talent Ops team, Kristen and Dina. They both joined Traversal for the same reason: the people. In fact, both were sold on the team over lunch during their interviews. Kristen has been in builder mode since her days at Pinterest and Grammarly, where she helped build recruiting and talent programs through major stages of growth. Dina spent three years in campus recruiting at Datadog before supporting startup operations at Partiful. Together, they’re helping us build the next generation of Traversers at an exciting stage. They’ve embraced the chance to shape how we hire, from the systems and metrics behind good decisions to the processes that help us get there. As Kristen puts it: “I like the puzzle of it, taking something messy and building the systems and structure that let a team scale without losing what made the team special in the first place.” They’ll also be the first to tell you that recruiting is a team sport; it takes a process built to surface the right signal, with a team that takes hiring seriously. That’s one of the things Dina appreciates about Traversal: “Everyone cares so much, and I know I can always rely on people to jump in and help, even when it’s not directly their responsibility. You really see that in recruiting. Everyone is invested in getting our processes right and giving candidates the best experience.” A few other things to know about our Talent Ops team: they’re big believers in coffee walks (even though Kristen doesn’t drink coffee), and around 11 a.m. every day, both sport baseball caps at their desks thanks to the skylight above them (shoutout to Ashby for Dina’s favorite swag and recruiting software). We’re hiring across the company and adding new roles every week. Check them out: https://lnkd.in/gQSFykiz

Traversal Workers spent months running on our own production before we brought them to customers. Hear our product team talk about the Friday afternoon incident that made them realize we had something special. More from Eric and Lyndon on the making of Workers to come!

The hardest part of building an AI SRE isn't the model. It's the architecture underneath it. Traversal's architecture is five proprietary layers, built for production at scale: 1️⃣ Agentless Data Capture™ — captures production data without new agents, sidecars, or schemas 2️⃣ Causal Indexer™ — distills massive telemetry while preserving causal structure 3️⃣ Knowledge Bank™ — captures operational knowledge 4️⃣ Production World Model™ — maintains a living model of the entire environment 5️⃣ Causal Search Engine™ — tests thousands of hypotheses in parallel across multiple hops to find root cause Read how the five layers work together: https://lnkd.in/eKSW_v2R

Traversal’s ORCA-bench is now on the Specialized Intelligence Index, Fireworks AI's new index for real-work benchmarks across industries. Its inclusion is external validation of what we set out to build: a rigorous benchmark for production root cause analysis. We built ORCA-bench with Columbia University and Cornell Tech because production root cause analysis lacked a benchmark that reflects what on-call actually looks like. So we recreated it: a live, OpenTelemetry-instrumented e-commerce app with 19 microservices across 13 languages, 50GB+ of telemetry collected over six days, and 1,079 tasks varying report specificity, detection delay, and concurrent failures. The results are a reality check: -- The strongest of five frontier models achieved just 25.3% RCA accuracy on medium-difficulty incidents -- That dropped to 10% when starting from a vague report like “users are reporting site issues” -- Removing source code access cost every model 9–16 percentage points of accuracy -- The weakest model proposed an implausible root cause roughly 40% of the time Understanding production takes much more than a frontier model. Full methodology, dataset, and leaderboard on SII: https://lnkd.in/gVFWbXUG

We've been busy this summer. Traversal Workers, Knowledge Bank™ 2.0, Microsoft Teams, MCP servers, faster processing, revamped Analytics, live incident transcripts, and more. Here are some of the highlights ↓

Coding agents understand your code. They don’t understand the production environment it’s about to run in. For months, customers have been using Traversal MCP to close that gap, bringing real topology, dependencies, recent changes, and system behavior directly into tools like Claude Code. That means Traversal can flag how a change could affect production before it ships, not just investigate what went wrong afterward. Most MCP servers provide access to an API. Traversal MCP provides access to a continuously updated model of your production environment. Read the blog we published back in March on how it works. Link in comments.

AI shouldn’t be a black box, especially in production. That’s why Knowledge Bank™ lets you see and interact with everything Traversal knows about your environment, review what it’s learned from investigations, and add the context that only exists in your team’s heads, all in one place. Traversal builds that understanding automatically. You don’t have to add anything, but when there’s context you want to contribute, you can.

A few scenes from summer at Traversal ☀️ Big wins, team breakfasts, baseball, Friday fro-yo, shuffleboard, watch parties, and much more. Look fun? Come join us. traversal.com/careers

Putting an AI SRE to work shouldn't be a project of its own. Traversal Workers scan your environment and deploy themselves where they're needed – one click, no custom instructions. They join the right channels immediately, and new ones as they appear. One console shows every Worker and what it's investigating. Customers have scanned, deployed, and seen value all in the same day. It’s that easy.

A busy fall ahead. Here are a few of the places you can find us: https://lnkd.in/gWG3Ngnw