Anthropic's CCO Wants "Very Nontraditional." Boolean Returns 1,700 Palantir Alumni.

On May 28, 2026, Bloomberg ran a feature on how Anthropic hires. The quote that should make every technical recruiter pause came from Chief Commercial Officer Paul Smith: "I'm looking for someone very nontraditional, like, very nontraditional." He went on to describe an ideal candidate who can immerse themselves in a customer's world and help executives figure out where AI actually fits. If you ran that brief through LinkedIn Recruiter this morning, you got back a Palantir alumni list. That is the opposite of what Smith said he wants.

This is the core sourcing problem at the frontier-model tier in 2026. The job specs are concrete. The deliverables are concrete. The trait Smith is actually buying for is unsourceable by title or company filter.

The title filter returns the average, not the edge

Run the obvious Boolean. "Forward Deployed Engineer" OR "Forward Deployed Software Engineer" OR "Applied AI Engineer," US only. You will land on roughly 1,700 people. Palantir Technologies dominates by a wide margin. The rest fragments across Modal Labs, Cursor, ElevenLabs, Scale AI, Pinterest, and a long tail of stealth and seed shops. Top metro is New York.

1,700

That list is a trap for two reasons. First, supply is laughably short relative to demand: forward deployed engineer job postings jumped more than 800% between January and September 2025. Second, the people inside that list are the least differentiated candidates available, because every recruiter at every AI lab is messaging them this week. If you hand a hiring manager the Palantir alumni dump, you have delivered the median, not the edge.

Smith's brief is explicitly the edge case. "Very nontraditional" is not a filter you can type. It is a judgment about whether a specific person can sit across from a general counsel at Freshfields, understand what an insurance underwriter at PwC actually does for ten hours a day, and translate that into an MCP server that ships. The title "Forward Deployed Engineer" is a proxy for that work. It is not the work.

What Anthropic is actually paying for

Read the public Anthropic FDE job spec on Greenhouse and the deliverables are unusually concrete: MCP servers, sub-agents, and agent skills used in production workflows, white-glove deployment support for strategic accounts, and codifying repeatable deployment patterns back to Product and Engineering. The requirement says 3+ years in a technical, customer-facing role such as Forward Deployed Engineer, or as a Software Engineer with consulting experience. It explicitly encourages former technical founders to apply.

Notice what that requirement does not say. It does not require a current FDE title. It does not require a Palantir, Scale, or Databricks badge. The "former technical founder" line is a deliberate door for people whose last two years look nothing like a standard enterprise GTM résumé.

Smith's public customer proof points reinforce this. He cites PwC cutting an insurance underwriting cycle from 10 weeks to 10 days with Claude. He cites Freshfields, Quinn Emanuel, Holland & Knight, and Accenture's in-house legal team all building on Claude. The FDE who lands those accounts is not the person who can spell "Salesforce MEDDIC." It is the person who can read a 60-page underwriting manual on a Tuesday and ship an eval harness for it by Friday.

The bar Smith sets for others is deliberately the inverse of his own résumé. </pull> That is worth sitting with. Smith himself came from ServiceNow, where he ran global customer and field operations, after Salesforce, after Microsoft. He is a 30-year enterprise-software operator. The CCO seat went to that pedigree. The roles under him will not. Recruiters who pattern-match the hiring manager's résumé onto the candidate brief will miss the actual instruction. ## The artifact replaces the résumé If title and company filters return the wrong pool, what replaces them? Artifacts. The work itself is increasingly public. A 2026 FDE candidate who fits Smith's brief usually has a trail: - A GitHub repo with a working MCP server for a specific vertical (a clinical notes parser, a contract redline tool, a claims triage agent). - A public Claude or OpenAI integration shipped under their own handle, not a company logo. - A conference talk or long-form blog post that demonstrates customer-domain fluency, not framework fluency. - A prior company they founded, often failed, that forced them into the customer's seat for 18 months. None of that is in a LinkedIn title field. All of it is on the open web, on GitHub, in YouTube transcripts, in Substack archives, and in conference programs. The sourcing job is to query that surface, not the résumé surface. This is the specific friction we built [Refolk](/) to remove. You describe the person in plain English, including the artifact you want them to have shipped, and you get a ranked shortlist across GitHub, LinkedIn, and the open web. "Find me people who shipped an MCP server for a legal or healthcare workflow in the last 12 months, regardless of current title." That is a sourceable query when the index includes artifacts. It is unsourceable in a Boolean editor.

refolk prompt: Find former technical founders now working as solutions architects or sales engineers who have shipped a public MCP server or Claude integration for legal, healthcare, or insurance workflows in the last 18 months. note: You get a ranked list with GitHub artifacts, prior company context, and current role pulled from LinkedIn and the open web, ordered by domain immersion evidence rather than title match. slug: vrrn96cxh2

## The pre-title arbitrage

The cleanest tactical takeaway from the Bloomberg piece is this: source pre-title.

The people doing FDE work in 2026 are often not titled FDE. They are titled Solutions Architect at a Series B vertical SaaS. They are titled Sales Engineer at a developer-tools company. They are titled Agent Engineer, Founding Engineer, or Member of Technical Staff at a 12-person startup. Peter Bayliss, the former Workday CTO, joined Anthropic as a Member of Technical Staff. The C-suite-to-IC archetype is real, and it does not surface in a title search.

The arbitrage move is to search for the *behavior* rather than the *label*:

1. **Customer-facing technical writing under their own name.** Substack, personal blog, company engineering blog with their byline. Bonus if the post explains a customer's domain back to them better than they could.
2. **Domain-specific OSS.** A repo whose README is written for radiologists, underwriters, litigators, or supply-chain planners, not for other engineers.
3. **Conference talks at vertical events.** Not AI Engineer Summit. HIMSS. Legalweek. InsureTech Connect. RSA. The person who chose to speak there understands the customer.
4. **Former founder, current IC.** Specifically people who founded and wound down a vertical AI startup in 2023 to 2025, and are now doing FDE-shaped work under a Solutions or Applied AI title at a larger lab.

Anjor Kanekar is the canonical example of what this looks like at the extreme. He spent seven years as a Palantir FDE and at one point worked on the final assembly line at Airbus alongside peers in airgapped environments. That is what "immerse themselves in a customer's world" means in practice. It is not a slide. It is showing up on the factory floor.

## The supply side is breaking, fast

Anthropic now has just over 3,000 employees per Live Data Technologies, with roughly 1,000 added since November 2025. The company is still hiring lawyers, accountants, finance professionals, salespeople, and marketers globally. The HR team has doubled because applications are overwhelming. London recruiter Jade Hussain reported 1,000+ LinkedIn connection requests and 200 messages within days of announcing her start, and later had to ask candidates to stop calling her personal phone.

```stat
number: 800%+
label: Growth in FDE job postings, Jan to Sept 2025
note: The title is hot. The pool is tiny. Every lab is fishing the same 1,700 profiles.
</stat>

The combination of $250K+ packages (some vacancies pay up to $850K), a sub-2,000-person titled US pool, and 800%+ posting growth is a textbook auction. The candidates inside the auction get the best comp. The candidates outside the auction get found by whoever sources artifact-first.

Comp anchors for context: Google Cloud FDE base sits at $127K to $183K plus equity. OpenAI mid-level FDE in San Francisco runs $220K to $280K with up to 50% travel. Anthropic clears those bands at the top. Interviewing.io reports the average candidate landing at Anthropic or OpenAI spends about $4,600 on interview prep. That is a candidate population that knows exactly what it is worth, which is the worst possible population to compete in by message volume.

The recruiter who wins here is the one who shows up in a former clinician's inbox saying "I read your RAG-for-radiology repo, your post about why oncology workflows break LLMs, and your talk at HIMSS, and I think you are the closest fit to a job spec I am working on at Anthropic." That message is not produced by a Boolean. It is produced by an index that knows about the repo, the post, and the talk.

This is where artifact-first sourcing actually changes the unit economics of nontraditional candidate sourcing. The reply rate on "I saw your work" is structurally higher than on "I saw your title." Refolk surfaces those artifacts as a first-class signal alongside LinkedIn, which is what makes "very nontraditional" briefs sourceable at all.

## What an artifact-first brief looks like

Translate Smith's "very nontraditional" into a brief your sourcing team can actually run:

- **Domain immersion evidence.** At least one of: a vertical OSS repo, a customer-domain blog series, a vertical conference talk, prior employment inside the target customer industry.
- **Shipped AI artifact in the last 18 months.** MCP server, Claude or GPT sub-agent, agent skill, eval harness, RAG system, in production or with public traction.
- **Customer-facing technical role, broadly defined.** Solutions Architect, Sales Engineer, Forward Deployed Engineer, Applied AI Engineer, Founding Engineer at a B2B startup, ex-consultant at a top firm with a technical track.
- **Negative signal.** No customer-domain artifact, no shipped AI artifact, pure platform engineering background with no GTM exposure. These are great engineers and wrong for this seat.

That brief is a search you can run today. It is not a Boolean. It is a description. Tools like Refolk exist because the description is now the right interface for sourcing AI talent at this tier, and the title field stopped being predictive sometime in late 2024.

Anthropic told you what they want in plain English. Source in plain English back.

## FAQ

### Why does Boolean fail specifically for Anthropic-style GTM roles?

Boolean filters on stated attributes: title, company, location, school. Smith's brief is an unstated attribute: the candidate's ability to immerse in a customer's world and translate AI capability into executive decisions. That trait correlates weakly with title and almost not at all with company logo. It correlates strongly with artifacts the candidate has shipped, which Boolean cannot see. The 1,700-person US FDE title pool is also small enough that every major AI lab is contacting it simultaneously, which means even when Boolean does return the right names, the messaging environment is saturated.

### Should I stop sourcing Palantir FDEs entirely?

No. Palantir alumni are real candidates and several have been excellent Anthropic hires. The point is that they should be one slice of your pipeline, not the entire pipeline. If your shortlist is 90% Palantir, you have under-indexed on the "very nontraditional" half of Smith's brief and you are bidding against every other lab for the same names. Mix in former technical founders, vertical-domain operators with AI artifacts, and IC roles at non-AI companies who ship Claude integrations on the side.

### How do I evaluate "domain immersion" without interviewing every candidate?

Look for evidence the candidate spent real time inside a customer's workflow before building software for it. Proxies include prior employment in the target vertical (clinician, lawyer, underwriter, supply-chain operator), public writing that explains the domain back to insiders, OSS where the README is written for non-engineers in that vertical, and conference talks at industry events rather than AI events. The Anjor Kanekar archetype, an FDE on a factory floor, is the high bar. Most strong candidates show at least two of the above.

### Does artifact-first sourcing work for non-engineering GTM roles too?

Yes, with a different artifact set. For AI go-to-market hiring outside engineering, the artifacts are customer-facing writing, podcast appearances, deal narratives published on LinkedIn, and product launches the candidate clearly drove. Smith is hiring marketers and salespeople under the same "nontraditional" bar, and the same logic applies: stated title is a weak signal, public output is a strong one. The query changes from "shipped an MCP server" to "wrote the launch narrative for an AI product that named customers cite," but the method is the same.