Cursor's 150-Person Roster vs Hiring Assistant 2: A Poach Map

LinkedIn shipped Hiring Assistant 2 at Talent Connect this week, the same week Anysphere's hiring process still forbids AI tools in first-round coding screens and still requires an 8 to 9 hour on-site build with the core team. Anysphere sits at roughly 150 employees, a $29.3B valuation as of November 2025, and $1B+ ARR that same month (reportedly $2B by February 2026). That combination means every Cursor engineer is now a named, finite, poachable target, and the agent LinkedIn just shipped is the exact wrong tool to find them.

Why Cursor's roster is the sharpest poach list in AI tooling right now

Because the team is small, revenue per head is historic, and the hiring filter actively selects against the keywords every agentic sourcer defaults to. At ~150 people and $2B ARR, Cursor is clearing roughly $13.3M per employee, so each departure is a measurable P&L event for Anysphere and a measurable capability gain for whoever lands them.

The counter-cultural piece matters more than the valuation. On Y Combinator's podcast, CEO Michael Truell said plainly: "We actually still interview people without allowing them to use AI, other than autocomplete, for first technical screens." He went further: "We've hired lots of people who are fantastic programmers who actually have no experience with AI tools. We would much rather hire those people and then teach them on the job."

That is a hiring criterion orthogonal to the LinkedIn signal graph. The result:

• A roster where many engineers do not list "LLM," "RAG," "LangChain," "vector," or "agent" on their profiles.
• A roster selected on raw programming taste rather than framework fluency.
• A roster that Hiring Assistant 2 and any other keyword-first automation will systematically underweight.

$13.3M

The anti-AI interview is a reverse filter for sourcers

The ban on AI in Cursor's first-round screen is the single most useful sourcing signal in AI tooling this quarter, because it tells you exactly which keywords will not surface this team. Any sourcer who leads with "LLM experience" or "agent framework" is searching in the wrong index.

Cursor's intake pipeline is biased toward programmers who can build cleanly from first principles, not candidates who have spent two years gluing SDK calls together. So when you reverse-engineer their likely roster:

1. GitHub signal matters more than LinkedIn signal. Commit history, language depth, and systems-level repos beat skill tags.
2. Compiler, runtime, database, and editor-internals contributors are overrepresented versus the general AI-tooling pool.
3. Competitive-programming history carries weight the LinkedIn graph does not encode.
4. The MIT cofounder lineage (Truell, Sualeh Asif, Aman Sanger, Arvid Lunnemark) biases hiring toward programs where "fantastic programmer who hasn't touched LLMs yet" is a real archetype.

This is the gap Refolk closes. You describe the person in plain English ("senior systems engineer, strong C++ or Rust, meaningful OSS contributions to a compiler or editor, no requirement for LLM experience") and get a ranked shortlist across GitHub, LinkedIn, and the open web, rather than a LinkedIn-only pull that misses half the actual roster shape.

The dataset: what the poach map actually looks like

Here is the comparable data behind building a named list against Cursor's ~150 engineers, pulled from Refolk's index and public reporting on Anysphere.

9,962

Two things to notice. First, the AI-lab concentration (16%) is a smaller share of the senior pool than most sourcers assume, which means the "poach from OpenAI and Anthropic" default is already crowded relative to the actual supply. Second, 36% of the top sample sits in San Francisco or the broader Bay Area, which is where Cursor's two-day on-site logistically selects for and where competing offers have to physically land.

Who already concentrates this profile

If you are either hiring against Cursor or sourcing into Cursor, these are the employers Refolk's index flags as currently concentrating the senior PyTorch + LLM profile. The list runs both directions: these companies feed Cursor's pipeline, and they are the most likely destinations for Cursor flight risk after the reported SpaceX/xAI transaction prints.

• Meta (top single employer in the sample)
• OpenAI
• Anthropic
• Palantir
• Reflection
• Mintlify
• Whop
• FurtherAI

Secondary pool worth mapping separately: ex-Graphite engineers now inside Anysphere via the acquisition reported above Graphite's $290M valuation. These are a cleanly named group whose equity vesting schedules and work scope changed overnight, which historically is when outreach lands.

Any Cursor-shaped search should also include contributors to open-source editors, LSPs, and code-navigation tooling. Those profiles rarely surface in LinkedIn Recruiter boolean strings but routinely show up on GitHub.

What LinkedIn Hiring Assistant 2 just did to Cursor's inboxes

Hiring Assistant 2 raised the noise floor against every finite roster, and Cursor's 150-person roster will feel it this quarter more than almost anyone. LinkedIn disclosed over 20,000 companies now use its agentic hiring solutions in year one, and that recruiters are 4x more likely to contact candidates surfaced by Hiring Assistant than those found through traditional methods.

Applied to a finite population:

• 150 engineers at Cursor will each field materially more generic agent-written outreach starting now.
• The announced reference customers (AMD, Cisco, Ochsner Health, Palo Alto Networks, Zillow) are the contrast class: companies now armed with agentic sourcing by default.
• Reply rates on template outreach to this profile will drop, not rise, as volume rises.

The wedge left is personalization grounded in signals the LinkedIn graph does not hold: a specific commit, a specific repo, a specific conference talk, a specific review on a specific PR. "I saw your work on X" lands only when X is real and specific.

The equity story just changed, and the window is now

The New York Times reported in April 2026 that SpaceX (through xAI) had optionally agreed to acquire Anysphere and Cursor for US$60B, which converts private, illiquid Cursor equity into something closer to a liquidity-eligible position. That historically weakens retention rather than strengthens it. If you are planning to poach Cursor engineers, the useful window is between the deal announcement and the close, not after.

Why this matters mechanically:

• Pre-liquidity, the comp floor you have to beat is base plus a highly illiquid equity grant valued at $808K to $1.1M+ total.
• Post-liquidity, the floor becomes base plus a tradable position, which is a different negotiation.
• During the window, cliff math, double-trigger acceleration, and refresh grants are all in flux, which is when engineers answer messages they would otherwise ignore.

Mass-automation sourcing is the exact wrong tool for a hiring filter designed to reject keywords.

The pitch that actually works against a $1M+ comp floor

Cash does not win here, scope does. Any Cursor engineer who completed the 8 to 9 hour on-site project day (part of the real codebase, a Slack channel for questions, a real feature shipped and presented to the team) has already demonstrated they will do a full day of unpaid, high-stakes work because the problem interests them. That is a behavioral tell. The recipient is motivated by problem and shipping surface area, not price.

Pitches that work on this profile:

1. A specific unsolved problem. Not "we're building the future of X." A specific technical gap you need them to own.
2. Shipping cadence. How often code lands in production, and how fast a decision gets made.
3. Scope of ownership. A real surface area, named. "Own our editor-side latency" beats "join our platform team."
4. A named founding peer. Who they would work next to, by name, with a link to that person's work.
5. Clear equity math. Strike, preference stack, and refresh policy, not a hand-wave.

Pitches that fail:

• "Competitive comp, equity, unlimited PTO."
• "We're a Series B / Series C AI company."
• "15-minute intro call with our recruiter."

The founders to displace in any narrative are the four MIT cofounders (Truell, Asif, Sanger, Lunnemark). Any pitch has to answer the implicit question: why leave the people you chose to grind the on-site for, for the people in this email.

A 10-step poach map against Anysphere

Here is the concrete sequence, built to work against a finite 150-person roster rather than a general AI-tooling market.

1. Build a named roster of ~150 current Cursor engineers from GitHub, LinkedIn, and conference talks.
2. Tag each by cofounder adjacency, join date, and prior employer (Graphite subset flagged separately).
3. Cross-reference against the 9,962-person Refolk index of senior PyTorch + LLM engineers to find overlap and gaps.
4. Pull parallel rosters at Meta, OpenAI, Anthropic, Palantir, Reflection, Mintlify, Whop, and FurtherAI as both source and destination maps.
5. Add an OSS-first signal layer: compiler, runtime, LSP, editor-internals contributors with no AI keywords, because that is where Cursor's actual hiring filter points.
6. Score for the SpaceX/xAI deal window: engineers with vesting cliffs in the next two quarters first.
7. Draft outreach grounded in a specific commit or review, not a title.
8. Lead with scope and a named peer, not comp.
9. Expect reply rates to compress as Hiring Assistant 2 volume lands this quarter.
10. Track response rates by signal source weekly, and prune the automated channels first.

Refolk handles steps 1, 3, 4, and 5 directly: you describe the person in plain English, and the signals LinkedIn Recruiter cannot see are already in the index.

FAQ

How many engineers does Cursor actually have?

Public estimates range from roughly 50 to 300 as of April 2026, with a working midpoint at ~150. Reporting from jobsbyculture.com cites a team around 50 at the $2B ARR milestone in February 2026 and a broader headcount footprint growing fast through the year. For sourcing purposes, treat the roster as finite and nameable: whether it is 150 or 250, it is small enough to build a named poach list against end to end.

Does Cursor really ban AI tools in interviews?

Yes, in first-round technical screens. CEO Michael Truell on Y Combinator's podcast: "We actually still interview people without allowing them to use AI, other than autocomplete, for first technical screens." Candidates who clear screens are invited to an 8 to 9 hour on-site project day where they build a real feature inside part of the Cursor codebase and present it to the team. The ban is a deliberate filter for raw programming taste, which is also why keyword-first sourcing tools underperform against this roster.

What total compensation do I have to beat to poach a Cursor engineer?

Per verified compensation reports, software engineers at Cursor typically clear $808K to $1.1M+ in total comp, including base and equity. That is the cash-equivalent floor, but the behavioral evidence (that candidates completed a two-day unpaid on-site) suggests cash alone will not move most of this roster. Scope, shipping cadence, named peers, and clear equity math matter more than the headline number.

Why won't LinkedIn Hiring Assistant 2 find these people?

Because Cursor's hiring criteria are orthogonal to the LinkedIn signal graph. The team deliberately hires "fantastic programmers who actually have no experience with AI tools," so a large share of the roster does not carry the LLM, RAG, LangChain, or agent keywords that Hiring Assistant 2 and similar agentic sourcers default to. GitHub, OSS contributions, and conference talks hold the signal; LinkedIn skill tags do not.