Instinct Hit $10B in 33 Days: The 40 Names to Source First

On Sep 28, 2026, Noah Shinn's Instinct confirmed a $1B Series C led by Sequoia, Benchmark, and Coatue at a $10B valuation, 33 days after a round that priced it at $2.5B. The company has 14 employees. If you are a recruiter staring at an empty LinkedIn search for "Instinct," you are looking in the wrong place.

The sourcing move here is not to poach from Instinct. There is nothing to poach yet. The move is to beat Shinn's own hiring list to the people he is about to call, and to pre-empt his competitors (Meta's Muse team, Sierra, every Benchmark-backed agent startup) before Sequoia-funded recruiters normalize comp across the whole personal-agent category.

Why the standard "ex-Instinct" Boolean is useless this quarter

The Instinct roster is too small and too new to source against directly. With roughly 14 total employees and under two months of public existence, there are maybe 8 to 10 engineers on LinkedIn with the company in their title, and most of those profiles are either stale or deliberately quiet.

The real target list is the three concentric rings around Shinn himself:

1. The Reflexion paper co-author tree. Six names, their students, and anyone who has cited the paper in the last 24 months with an agent-related repo.
2. Sierra alumni. Shinn's former employer. Sequoia-backed (same lead as Instinct's Series C), warm-intro-rich.
3. The voice/telephony primitive layer. LiveKit, Vapi, Retell AI, and Twilio-heavy agent builders. This is Instinct's actual differentiator and its actual bottleneck.

The 33-day math nobody has priced in yet

Instinct raised roughly $1.35B total across its Series B and Series C and still has 14 employees, which works out to about $96M of capital per employee. That is the real recruiting weapon, and it is why every competing offer this quarter is about to feel underwater.

$96M

Here is what the research note's numbers look like side by side.

Two numbers in that table do the work. The 398 US "AI agent" engineers is roughly the entire addressable pool Shinn is about to hire from, and about 40% of them sit in the Bay Area. The 947 LiveKit engineers worldwide is the ceiling on anyone who can actually ship Instinct's phone-call feature at scale, and the top region for that skill is Bengaluru, not San Francisco.

If you want to know why Shinn can plausibly hire 150 to 300 people before diluting meaningfully, the answer is in the capital-per-head figure. If you want to know who is going to feel the squeeze first, the answer is the LiveKit row.

Ring 1: the Reflexion co-author tree

Start with the six names on the 2023 Reflexion paper. That is Shinn's actual professional graph, in print, with GitHub handles attached.

• Federico Cassano. Reflexion co-author, Northeastern, codegen and agent researcher. The highest-signal single name on the list and the one every competent AI recruiter should already have in a sequence.
• Edward Berman. Reflexion co-author, Northeastern undergrad cohort with Shinn. Likely still early career and realistically poachable.
• Ashwin Gopinath. Then at MIT. Builds entire lab pipelines, which means his students and postdocs are a second-order list worth 20 to 30 names on their own.
• Karthik Narasimhan. Princeton faculty, agent and RL research. Unpoachable himself, but his lab is a feeder.
• Shunyu Yao. Now Tencent's chief AI scientist and author of ReAct. The single most-cited "agent" researcher alive. Also unpoachable, also a feeder.
• Noah Shinn. Founder of Instinct. Obviously not for sale.

The move on this ring is not to cold-message the six co-authors. The move is to pull every US-based PhD who has cited Reflexion since early 2024 and lists at least one agent-related repo on GitHub. That is roughly where Shinn's own internal shortlist starts, and it is where you want to be two weeks ahead of him.

Why Sierra alumni outrank ex-OpenAI alumni here

Sierra is Shinn's former employer and the strongest adjacency pool, which almost nobody is targeting because the default "ex-OpenAI, ex-Anthropic" Boolean sucks all the oxygen out of agent sourcing. Sierra is co-founded by Bret Taylor (chairman of OpenAI) and backed by Sequoia, which also leads Instinct's Series C. Those facts together mean warm intros already exist, comp expectations are already calibrated to agent-startup equity, and Shinn personally knows a meaningful share of the engineering org.

If you are only running "ex-OpenAI" searches, you are competing with every other agent recruiter in the Bay Area. Sierra is quieter, denser, and closer to Shinn's actual network.

Describing that pool as a Boolean is painful. Describing it in plain English to Refolk ("Sierra engineers with 18+ months tenure and an agent-shaped title, US-based") returns the ranked list in seconds.

Ring 2: the Muse mirror and the Benchmark portfolio

Meta's AI chief Alexandr Wang publicly compared Instinct to Meta's Muse product, which turns the Muse engineering team at Meta Superintelligence Labs into the direct mirror pool. Those engineers are pre-qualified on comp expectations because of Meta's recent compensation inflation, and they are building the exact product shape Instinct is trying to beat.

The second half of this ring is the Benchmark and Index Ventures agent portfolio. Benchmark co-led the $250M Series B with Index in August and returned for the Series C. Any agent-adjacent portfolio company at either firm now has an awkward conflict dynamic, and some of their engineers will take the call for that reason alone.

Ring 3: the LiveKit and voice-telephony squeeze

Voice and telephony engineers, not frontier AI researchers, are the actual bottleneck for Instinct and its competitors right now. Instinct's product differentiator is an agent that can text or call on a user's behalf, which means the stack is LiveKit, WebRTC, Twilio, and the thin layer of people who have shipped these primitives against an LLM at latency.

Refolk's index shows 947 engineers globally who list LiveKit as a skill. The top region is Bengaluru, not a US city. That means the US sliver is small, premium-priced, and about to get hammered by three or four well-funded buyers at once.

The role families to target here are not "AI research." They are:

• Voice infrastructure engineers at Vapi and Retell AI.
• Senior backend engineers at Twilio who have touched the Programmable Voice stack in the last two years.
• Early engineers at any YC voice-agent startup from the recent batches.
• LiveKit's own alumni (the company is small, so this is a double-digit list, not a hundred-plus one).

Running these searches as nested Booleans across LinkedIn, GitHub, and company directories is a two-week project. Running "US LiveKit engineers who have shipped a production voice agent, not currently at Vapi or Retell" through Refolk is a one-sentence prompt.

What "invite-only, no metrics" tells you about the culture Shinn is hiring for

Instinct's launch posture is a culture signal, not just a PR choice. The announcement disclosed no user count, no revenue, no pricing, and no date for opening the invite-only service to everyone. That tells you Shinn is optimizing for a tiny closed user base with absurd per-user intimacy, which rules out most scale-stage hires.

The comparable engineering cultures are not OpenAI or Anthropic. They are early Superhuman, early Linear, and early Arc Browser: obsessive latency, product-engineer-as-designer, opinionated defaults, high-density taste.

Target those alumni networks the same way you target the Reflexion graph. The resume signal is "shipped a product that users describe as magic to 10,000 or fewer users."

Shinn is hiring taste, not scale. Target the people who built the first 10,000-user versions of Superhuman, Linear, and Arc.

The 60-to-90-day window and how to actually run the play

The sourcing window closes when Sequoia's in-house recruiting team spins up a dedicated pod on Instinct, which historically happens 45 to 75 days after a round of this size. Call it a 60-to-90-day window from Sep 28 to get in front of candidates before comp conversations get normalized across the category.

A reasonable sequencing plan looks like this:

1. Week 1. Pull the Reflexion citation graph (24-month window, US-based, agent repo filter). Target: 40 names.
2. Week 2. Layer in Sierra alumni with 18+ months of tenure and an agent-shaped title. Target: 30 names.
3. Week 3. Pull the LiveKit, Vapi, Retell AI, and Twilio Voice list, US only, with at least one production voice-agent ship in the last 12 months. Target: 50 names.
4. Week 4. Add the Muse mirror pool and the Superhuman/Linear/Arc culture tier. Target: 40 names.
5. Weeks 5 to 8. Sequence, book, close.

FAQ

Who are the Reflexion paper co-authors and why do they matter for Instinct sourcing?

The six co-authors are Noah Shinn, Federico Cassano, Edward Berman, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. They matter because Shinn's own hiring shortlist almost certainly starts with his co-authors, their students, and the researchers who have cited the paper since 2024. Mapping that citation graph gives you the same list Shinn is about to work, two weeks earlier.

Can I actually poach engineers from Instinct right now?

Not really. With 14 total employees and under two months of public existence, the "ex-Instinct" pool is effectively empty and the few people on it just took a 4x markup on their equity. The productive play is poaching into Instinct's competitors (Sierra, Meta Muse, every Benchmark agent portfolio company) and pre-empting Shinn's inbound list before Sequoia's recruiters move.

Why focus on LiveKit engineers instead of AI researchers?

Because voice and telephony are Instinct's actual product bottleneck, not frontier research. The global LiveKit pool is 947 engineers and the US sliver is small, which means three or four well-funded voice-agent startups are about to compete for the same couple hundred people. That is where the price action happens first.

How is Refolk different from running these searches on LinkedIn Recruiter and GitHub?

LinkedIn Recruiter cannot filter by "cited this paper" or "shipped a production voice agent in the last 12 months," and GitHub cannot filter by current employer or title. Refolk takes a plain-English description of the person you want and runs it across LinkedIn, GitHub, and the open web as one query, which is the only way to assemble lists like the Reflexion citation graph without burning two weeks of sourcer time.