Meta's TBD Labs Runs on 32 Names. Your Niche Is Probably Smaller.

If you run in-house recruiting at a company that will never write a $100 million signing bonus, Meta's TBD Labs looks like a story about money. It isn't. It's a story about method, and the method is yours to copy.

TheStreet's recent breakdown of Meta's grab of Chirantan "CJ" Desai, out of the MongoDB CEO chair in under 11 months and into Meta as Chief Enterprise Platform Officer, framed modern executive poaching through Cowen Partners as "highly targeted" work, a different discipline from applications and referrals. The budget isn't the point. The list is.

Meta's TBD Labs isn't hiring, it's running a list

TBD Labs is a roughly 50-person unit inside Meta Superintelligence Labs that operates as a named-target sourcing shop, not a hiring funnel. Mark Zuckerberg has described it as a boat where every member has to row. There is no inbound. There is a list of people Meta wants, and a sequence of closes against that list.

The CJ Desai hire is the shape of it:

• Sitting CEO at a public company (MongoDB).
• Less than 11 months into the job.
• Pulled "effective immediately" into a newly scoped C-suite role at Meta.

That is not a req. That is a one-person search with an opening offer and a decision window measured in days. Andrew Tulloch, co-founder of Mira Murati's Thinking Machines Lab, left for Meta in October on a package reportedly worth $1.5 billion over six years. Sam Altman has conceded that signing bonuses of up to $100 million have been dangled at frontier researchers. Mark Chen, OpenAI's chief scientist, compared the raids to "someone breaking into our home."

Strip the dollar figures and the method is identify, research, pitch. That is the part you can steal.

The pool is smaller than the headcount

In Refolk's index of professional profiles, there are only 32 US senior or director-level people whose current role explicitly centers on "LLM" or "foundation model." TBD Labs itself has about 50 members. The targetable pool outside Meta, for that exact seniority and keyword slice, is smaller than the team doing the targeting.

32

This is the point most sourcing conversations miss. People talk about "the AI talent pool" as if it were a labor market. For the roles TBD Labs is actually filling, it is a Rolodex. The top employers in that 32-person set are Apple with 5, Amazon with 4, and Google with 2. You could print the list, call it this quarter's target set, and have time left over for lunch.

Run the same math on your own niche: a Rust distributed-systems hire, a payments compliance lead, a staff ML engineer with recsys experience. The number is almost always under 200. There is no pipeline to build. There is only a list.

Why title filters miss the real pool

Widen the net to Research Scientist, AI Research Scientist, and Member of Technical Staff with LLM plus Transformers skills, and Refolk's index returns 154 US profiles. Of the top 25 titles in that set, 19 are "Member of Technical Staff." Anthropic sits at the top of the employer distribution with 3.

Member of Technical Staff is the frontier-AI tell. OpenAI and Anthropic use the title to flatten hierarchy and, functionally, to obscure seniority from recruiters trained to filter on "Senior" or "Staff." If your LinkedIn Recruiter boolean leads with a seniority qualifier, you are deleting the exact people you want. The real query is keyword plus employer plus tenure, which is the kind of description a plain-English tool like Refolk handles natively: you say "members of technical staff at Anthropic or OpenAI who have been there 18 months and work on post-training" and get a ranked shortlist, not a seniority-filtered void.

Named-person recruiting, in four moves

The TBD Labs playbook compresses to four moves, and all four port to a $90K base, not just a $1.5B package.

1. Define the list, not the req. Write the names before you write the job description. If you cannot name 20 candidates, your spec is wrong, not the market.
2. Research the pain, not the resume. Cowen Partners' phrasing in TheStreet: "actively identify high-value leaders, research their pain points, and deliver tailored pitches designed to resonate." Pain points are things like a product line being deprioritized, a boss change, a reporting-line reorg, a public down round.
3. Pitch the specific swap. The pitch names the thing they lose, the thing they gain, and the window. CJ Desai's pitch was almost certainly "we are building the enterprise AI distribution MongoDB can't fund in two years."
4. Compress the decision window. Mid-market recruiters cannot match $100M, but they can go from first message to signed offer in two weeks while LinkedIn Recruiter users are still waiting for an InMail reply.

Where mid-market actually beats Meta

Meta is slow in the places you are fast. Legal review on a Chief Enterprise Platform Officer compensation package is not a two-day exercise. Internal politics inside a company Meta's size slow every bespoke offer. A 40-person company with a founder on the pitch call can close a staff engineer in nine days.

Mid-market recruiters cannot match a hundred million dollars, but they can compress a decision window from months to days.

Speed is the actual competitive variable. The CJ Desai hook in Rob Lenihan's TheStreet piece was speed, not money: less than 11 months as CEO, out "effective immediately." If your pitch is tailored and your decision window is seven days, you are already competing with Meta on the only axis that matters to a person who has three other offers.

The 696-a-day layoff flood is noise, not signal

The tech-layoff count is useless as a sourcing input for frontier roles because the signal-to-noise ratio has collapsed. So far in 2026, TrueUp has logged 190,077 people cut across 623 tech-company layoff events, a rate of 696 per day. That is actually higher than 2025's 674-per-day pace, even though cumulative volume is lower. TrueUp projects the 2026 ceiling at 370,000, significantly exceeding the previous two years.

The underlying cause is the same reason targeted sourcing matters: Meta, Atlassian, and Cloudflare are cutting to fund AI, which is a long-term structural shift. Meta itself laid off roughly 600 employees in its AI unit in October, and TBD Labs was explicitly protected from the cut.

Divide 190,077 by the 32-person named-target pool and you get a ratio of roughly 5,940 to 1 between the inbound flood and the actual list. Posting a req gets you unusable volume. Sorting through 400 applications is not sourcing, it is customer-service triage with a shortlist of zero at the end of it.

The layoff count is the reason targeted sourcing works right now, not the reason to abandon it. The people you want are not laid off. They are sitting at Anthropic, Apple, Google, and a handful of labs you can name, and they will not apply.

The two signals that make a named-target list

A named-target list is built from two signals: pain and proximity. Pain is why they will take the call. Proximity is how you find them in the first place.

Pain signals, in rough order of how reliably they open a reply:

• A direct boss change in the last 90 days (see: Yann LeCun's November 2025 departure after being told to report to Alexandr Wang, after 12 years as Meta's chief AI scientist).
• A reorg that moved them under a function they did not sign up for.
• A product line being deprioritized or spun down.
• A down round, a missed earnings call, a public RIF announcement.
• A tenure cliff: 11 months (CJ Desai), 18 months, 4 years.
• Co-founder or VP peer departures in the same quarter.

Proximity signals are what actually get them on the list:

• A GitHub commit history on repos adjacent to your stack.
• Co-authorship on a paper from the lab you are hiring against.
• A conference talk at an industry-relevant venue.
• Current employer plus a specific keyword in their bio or job title.
• A tenure window that implies vest cliff timing.

Most recruiting tools make you pick one signal at a time, which is why "sourcing" collapses into "LinkedIn boolean plus hope." The gap Refolk closes is combining them in one plain-English query: you describe the pain plus the proximity in a sentence and get a ranked list across GitHub, LinkedIn, and the open web, without stitching five tools together.

What happens after you name the list

The entire executive poaching strategy, once the list exists, is a research sprint and a sequencing decision. You will close roughly one in six to one in ten named targets at the frontier tier, which means a 20-person list produces two to three serious conversations and one hire. That is a successful quarter.

The order matters more than people expect:

1. Approach the second-best name first. You learn what the pitch is missing without burning the top target.
2. Interview their current pain in message two, not message one. Message one gets the reply. Message two earns the call.
3. Introduce the founder or hiring principal by message three. Recruiter-only sequences stall at reply four.
4. Make the offer in person or on video, never by email. Billion-dollar packages get sketched on whiteboards for a reason.
5. Give them a 72-hour decision window with a reason. The reason is almost always a board meeting, a kickoff, or another candidate.

The second-order effect of every big poach is that it creates new named targets. The incumbents the hire displaces, like LeCun after Wang's arrival, become next quarter's list. If you are running a mid-market search in AI infra, data platforms, or security, your target list this quarter is partly a function of who Meta hired last quarter, and the research work is tracing those shockwaves across the org charts you already know.

That is what named-person recruiting actually is. Not a bigger funnel. Not a smarter ATS. A list of people you want, a reason each one would take the call, and the discipline to work through them in order.

FAQ

How do I know if my role needs a named-target sourcing approach instead of a req?

Try to list 20 real candidates by name without opening LinkedIn. If you cannot, the pool is probably small enough that posting a job will drown you in unqualified applications and miss the people you want. The rule of thumb from Refolk's index: if the specific-skill-plus-seniority intersection returns fewer than 200 US profiles, treat it as a list, not a market. Niches like foundation-model research, cleared SOC leadership, or Rust distributed systems almost always land there.

What does Meta's TBD Labs actually do differently from normal Meta hiring?

TBD Labs operates as a protected, roughly 50-person unit inside Meta Superintelligence Labs that was explicitly excluded from the 600-person AI layoff Meta announced in October. It runs on named-target sourcing against a short list of frontier researchers, and the hires come with bespoke compensation packages negotiated outside the standard leveling grid. Normal Meta hiring is still a req-and-ATS funnel; TBD Labs is a search firm wearing a Meta badge.

Can mid-market companies really copy this without the signing bonus budget?

Yes, because the budget is not where the method lives. The method is identify, research, pitch, and the compounding advantage is speed. A 40-person company with the founder on the pitch call can close a staff engineer in nine days while a Meta offer waits two weeks for legal review. Match the discipline of the four moves (list before req, pain over resume, specific swap, compressed window) and you win the targets who care about proximity to the problem more than they care about the top line on an offer letter.

Why do title filters miss frontier AI candidates?

Because OpenAI and Anthropic use "Member of Technical Staff" as the default title across seniorities, which flattens hierarchy and defeats recruiter booleans that lead with "Senior," "Staff," or "Principal." In Refolk's index, 19 of the top 25 titles in the LLM-plus-Transformers research-scientist segment were Member of Technical Staff. Query on employer plus skill plus tenure instead, and you surface the exact people the seniority filter was hiding.