# Traversal's $120M AI SRE Needs Causal ML. Only 1 US Profile Has Both

_Traversal raised $120M to hire causal ML and reliability talent. Keyword search finds one US match. Source from papers, OTel commits and postmortems instead._

- Canonical URL: https://www.refolk.ai/blog/traversal-ai-sre-hire-causal-ml-reliability-engineers
- Published: 2026-11-21
- Reading time: 9 min
- Keywords: hire site reliability engineers, AI SRE engineer jobs, causal machine learning engineer, how to source SRE candidates, incident response engineer hiring

If you're hiring engineers who can apply causal ML to production incidents, the obvious searches fail right away. Search "causal inference" and you get founders and professors. Search "SRE" and you get reliability engineers at Google and JPMorganChase who never mention causal methods. In Refolk's index, those two US pools share exactly one person. Demand for that person just went up: Traversal raised a $120M Series B at a $1B valuation on November 3, 2026, and said the money goes to hiring across research, engineering and go-to-market.

## What is Traversal hiring for, and why is it hard to fill?

Traversal needs engineers and researchers who can apply causal machine learning to live production systems. Almost nobody puts that combination on a profile.

An **AI SRE** is software that does work a site reliability engineer would otherwise do: it watches production, triages incidents and finds the root cause. **Causal machine learning** is the branch of ML that estimates cause and effect, not just correlation, and root-cause analysis is a cause-and-effect question. Traversal's agents use causal ML to map production systems, test thousands of hypotheses in parallel and find the root cause of an incident.

The round details:

- **Amount:** $120M Series B at a $1B valuation, announced November 3, 2026
- **Leads:** Kleiner Perkins, Redpoint Ventures and Khosla Ventures, with significant participation from Sequoia and Battery Ventures
- **Total raised:** $173M
- **Usage:** monthly investigations grew from 513 to nearly 73,000 during 2026, more than 100-fold
- **Hiring plan:** research, engineering and go-to-market

The founders explain the hiring profile. Anish Agarwal and Raj Agrawal did their PhDs at MIT, Raaz Dwivedi did his at UC Berkeley, and Ahmed Lone came from quant trading at Citadel Securities. That's an academic team building for on-call engineers. The best causal machine learning engineer for this company knows the math and has also been paged at 3 a.m.

## Why does a "causal inference" search return economists?

In Refolk's index, most people who tag "Causal Inference" are founders, professors and product leaders, not systems engineers. Run the matching SRE search and you get a different crowd that never mentions causal methods.

| US pool in Refolk's index | Count |
|---|---|
| SRE-titled profiles | 7,206 |
| "Causal Inference" skill profiles | 9,224 |
| Causal Inference skill AND an SRE, Production, Observability or Infrastructure title | 1 |
| Causal pool vs SRE pool | 1.28x |
| Overlap as a share of the SRE pool | ~0.014% (1 in 7,206) |

The causal pool is bigger than the SRE pool, so supply isn't the problem. The problem is who's in it. In a sample of the causal pool, the top titles were CEO or founder, assistant professor, VP of product and applied AI engineer. None were reliability roles. On the SRE side, the top employers in the sample were Google, JPMorganChase, Apple and Morgan Stanley, so banks show up right alongside big tech.

> **1** US profiles in Refolk's index with the Causal Inference skill and a reliability title
>
> The one match is a Production Engineer at Meta, found among more than 16,000 profiles across both pools.

A caveat on the breakdowns: the title and employer mixes come from 25-profile samples, so read them as directional. The headline counts are full totals.

## Why can't Boolean AND search find this skill mix?

Boolean search only finds what people write on their profiles, and neither group writes the other group's words. The skill exists. It just isn't tagged.

The cause is how people label themselves. An SRE fills the skills section with tools like Kubernetes and Terraform. A causal researcher lists research titles and methods. An infrastructure engineer who builds dependency graphs to trace failures is doing causal reasoning every week, but has no reason to call it "causal inference." The economist who does call it that has never touched a pager.

Look at the one match, too. It's a *Production Engineer*, not an SRE. A search limited to the SRE title would have returned zero. Even a two-keyword AND search only works if you also cover title variants like Production Engineer, Observability Engineer and Infrastructure Engineer.

> The skill is real but untagged. Boolean search can only find what people chose to write down.

This is why plain-English search works better than keyword strings here. With [Refolk](https://www.refolk.ai/), you describe the person ("an infrastructure engineer who has written root-cause postmortems and published on causal or graph methods") and the search reads profiles, repos and papers together. You aren't relying on someone having typed both phrases into a skills box.

## Where does the causal ML and reliability overlap actually show up?

It shows up in things people produce: co-authored papers, OpenTelemetry commits, postmortems and course rosters. Each one is evidence of work, not a self-applied label.

### Papers that pair causal methods with systems co-authors

Co-authorship is the strongest signal because the paper itself joins the two skills. ORCA-bench is the template. Traversal built it with Cornell Tech and Columbia: 1,079 root-cause tasks, where the best of five frontier agents reached only 25.3% accuracy on medium-difficulty incidents and fell to 10% when the incident reports were vague.

The author list on a paper like that is a sourcing list. Anish Agarwal works on causal ML as a Columbia professor, and Raaz Dwivedi is Cornell Tech faculty. Their co-authors, students and collaborators sit in exactly the overlap you need. Look for papers that apply causal or graph methods to traces, logs, microservice dependencies or incident data, then check which authors have industry systems jobs.

### The OpenTelemetry contributor graph

**OpenTelemetry** (OTel) is the open-source standard for collecting traces, metrics and logs from production software. Resolve AI's founders, Spiros Xanthos and Mayank Agarwal, co-created it, and their earlier companies sold to Splunk and VMware. That makes the OTel contributor graph a map of who a direct competitor already knows. It's also where engineers who think about how telemetry is structured, the raw input to any causal model of a system, do their work.

Weight merged commits over stars or issue comments. [Merged PRs per repo](https://www.refolk.ai/blog/github-pr-caps-merged-prs-per-repo-sourcing) are the trust signal maintainers use themselves, so use the same signal when you rank contributors.

### Postmortem authorship

A **postmortem** is the written review of a production incident: what broke, what caused it and what changes. Writing one means someone did the root-cause reasoning Traversal is trying to automate. That tells you more than an SRE title does.

For incident response engineer hiring, search public engineering blogs and status-page writeups for named authors, and check incident-tooling repos for commits to runbook automation, alert routing or timeline tooling.

### Course rosters and TA lists

Traversal's first employee, Matthew Schoenbauer, started as a student in Agarwal's causal-AI class. Courses that teach causal methods to engineers, plus their TA lists and final-project showcases, produce people who learned the theory with systems examples.

> **Example search:** "Find US infrastructure or production engineers who have co-authored a paper on causal or graph-based root-cause analysis, or merged commits to OpenTelemetry." - [run it on Refolk](https://www.refolk.ai/s/q9kbmswd48)
>
> You get a ranked shortlist that links each person to the specific paper, repo or postmortem that qualifies them, across GitHub, LinkedIn and the open web.

## Who else is competing for the same engineers?

At least four well-funded AI SRE and observability companies are hiring from this small pool, so speed matters more than usual.

| Company | 2026 event | Value |
|---|---|---|
| Traversal | Series B, Nov 3 | $1B valuation |
| Resolve AI | Valuation step-up, April | $1.5B |
| Dash0 | $110M raise, March | $1B valuation |
| Observe | Acquired by Snowflake, January | ~$1B |

Each company hires the same profile: observability depth plus ML judgment. Resolve AI starts with an advantage in the OTel network. Traversal starts with one in the academic network. A candidate with a strong OTel commit history and a co-authored causal paper is the person every company on this list wants. Before you open, check current [engineering salaries in the United States](https://www.refolk.ai/jobs/salaries/engineering/in/united-states), because a $1B-valuation competitor's offer is the number you'll be compared to.

The live postings below show who is hiring for reliability work right now.

> **Site reliability roles open right now** - [see the open roles](https://www.refolk.ai/jobs/search?q=reliability&family=engineering)
>
> Live postings with reliability in the title, refreshed from the job board.

## Which layoffs actually free up this talent?

Dropbox is the better source for senior infrastructure engineers this quarter, not PagerDuty, even though PagerDuty is the bigger name in incident response. You can see why in who each company cut.

- **PagerDuty** filed an 8-K on August 27, 2026 to cut about 15% of staff. It reported 1,155 employees as of January 31, 2026, so that's roughly 172 people. The cuts fall primarily on non-customer-facing roles, and the company protected product teams.
- **Dropbox** is laying off about 9% of its workforce, roughly 200 people, mainly in engineering, product, design and G&A. CTO Ali Dasdan steps down on November 18, 2026.

That's roughly 372 people across the two companies, but they aren't equally useful to you. PagerDuty protected the teams closest to the product. Dropbox cut engineering directly, and the cuts came with its CTO leaving. The [earlier post on Dropbox's CTO exit](https://www.refolk.ai/blog/dropbox-cto-exit-infrastructure-engineers-recruiting) covers that team in more detail. Either way, find these engineers by what they built, not by the label on their profile. To turn a filing into named people, the [layoff-filing teardown](https://www.refolk.ai/guides/layoff-filing-team-outreach-teardown) walks from one notice to a dated, team-level slate.

## How to source SRE candidates with causal ML depth

Start with artifacts, confirm production experience second, and only then check titles. If you need to hire site reliability engineers who can also reason causally, this order puts the strongest evidence first.

1. **Build the paper list.** Start from ORCA-bench and similar root-cause papers. Pull every author, then flag the ones with industry infrastructure roles.
2. **Pull the OTel contributor graph.** Rank by merged commits in the last 12 to 24 months. Remove anyone already at Resolve AI, Dash0 or Traversal.
3. **Find postmortem authors.** Collect named authors from public incident reviews and from commits to incident-tooling repos.
4. **Widen the titles.** Search Production Engineer, Observability Engineer and Infrastructure Engineer alongside SRE. The one index match had a Production Engineer title.
5. **Layer in the layoff pool.** Check Dropbox engineering alumni against steps 1 to 3 before you look at the PagerDuty list.
6. **Cross-reference.** The best candidates turn up in two of these sources. One paper plus one postmortem beats ten keyword matches.

Done by hand, this is slow. A keyword search that returns one profile means you have to build every candidate yourself: read the paper, find the author's GitHub, confirm the employer, find a contact route. For a founder who also has to ship product, that research eats the time meant for outreach, while Resolve AI, Dash0 and Traversal itself work the same short list.

## FAQ

### How many US profiles combine causal inference and SRE experience?

In Refolk's index, one. There are 7,206 US profiles with an SRE title and 9,224 with the "Causal Inference" skill. When you require the skill together with an SRE, Production Engineer, Observability Engineer or Infrastructure Engineer title, one profile remains: a Production Engineer at Meta. The skill mix exists. It just isn't labeled, so you have to find it through papers, commits and postmortems.

### What should a causal machine learning engineer's background show for an AI SRE role?

Look for evidence of both halves. On the causal side, that means a paper, thesis or project that applies causal or graph methods. On the reliability side, it means postmortems written, incident tooling shipped or OpenTelemetry commits merged. Traversal's own team mixes academia (MIT, Berkeley, Columbia, Cornell Tech) with quant trading (Citadel), so don't screen out researchers who have production exposure.

### Are PagerDuty's layoffs a good source for incident response engineer hiring?

They help less than the name suggests. PagerDuty's roughly 15% cut fell primarily on non-customer-facing roles, and product teams were protected. Dropbox's roughly 200 cuts hit engineering directly, and the CTO leaves November 18, 2026, so Dropbox is the likelier source of senior infrastructure engineers this quarter.

---

_From the Refolk blog. More posts: https://www.refolk.ai/blog/index.md_
