Instinct's $714M Per Head: The Sierra Alumni Pool Is 21 Deep

On September 28, 2026, Instinct closed a $1B Series C at a $10B valuation with 14 employees on the payroll, which pencils out to roughly $714M of enterprise value per head. The founder, Noah Shinn, is 23, a Northeastern dropout, and spent about 2.5 years as one of Sierra's first research scientists before leaving to start the company. Every seed fund, every Series A board, and every incumbent consumer platform is now trying to clone that archetype, and the pool of engineers who can plausibly supply it is smaller than most recruiters think.

The number that broke this week's sourcing plans

Instinct is worth 35.7x more per employee than Sierra, the company its founder came from. That multiple is why the raid is happening, and why it will keep happening through Q1 2027.

Sierra, Bret Taylor and Clay Bavor's agent platform, raised a $950M Series E at $15.8B in May 2026 and has 789 employees across nine offices. Divide and you get $20.0M per head. Instinct quadrupled its valuation in a single month, going from $2.5B at the Series B to $10B on September 28, with Sequoia, Benchmark, and Coatue in the round. Fourteen people, $1B in new primary.

$714M

Any Sierra engineer who does the napkin math on four years of RSU vesting against a founding-engineer slice of a 14-person company has already done it. The question is where that engineer goes next and who gets to them first.

The Sierra alumni pool is 21 engineers deep, not 789

In Refolk's index of U.S. professional profiles, the Sierra engineering alumni and current-staff pool with U.S. engineering titles is 21 people. Of those, 14 sit in San Francisco proper. That is the raid-radius.

Sierra has 789 total employees globally, so the gap between the headline org chart and the actually-raidable American engineering pool is roughly 38x. The reasons are mechanical:

• Sierra's product and GTM organizations do not translate into founding-engineer resumes for a consumer-agent startup. The relevant pedigree is the research and infra core, not the solution architects staffing enterprise rollouts.
• Sierra runs an in-person requirement that pre-filters the pool toward Bay Area residents. Two-thirds of the raidable engineering pool sits inside one zip code cluster.
• The non-U.S. offices in London, Singapore, Tokyo, Paris, Madrid, and Toronto are mostly deployment and customer engineering, not the research-adjacent profiles that get Instinct-style term sheets.

The practical consequence: when a founder says "I want to hire Sierra engineers for my consumer agent startup," they are describing a list of 21 names. That is not a metaphor. It is the total addressable list, and 14 of them share a commute.

Why Sierra DNA is suddenly consumer DNA

Sierra built an enterprise customer-service agent business. Instinct is an invite-only consumer transaction agent running above $1B of annualized volume with zero marketing spend. The bridge between the two is not product strategy. It is evaluation infrastructure.

Shinn co-wrote Reflexion as the first author at 20 (91% on HumanEval versus GPT-4's 80%, published at NeurIPS) and co-built τ-bench at Sierra with Shunyu Yao. Both are evaluation and self-correction primitives, not pipelines. His papers now carry 10,000+ citations. The reason Instinct's product retains 80% of users who hand over a credit card inside three weeks, with 40% of all users doing so, is that the team ships with task-completion rate as the dependent variable and runs offline replays against real failure sets before any prompt change ships.

What this means for your intake

Recruiters chasing "ex-Sierra engineers who built retrieval pipelines" will overpay for the wrong profile. The premium sits on:

• Published or shipped work on agent evaluation (τ-bench and adjacent harnesses).
• Experience owning a regression set derived from production failures, not synthetic prompts.
• Self-correction and reflection loops at inference time (Reflexion-style architectures).
• Comfort with offline replay as the gating artifact before any model or prompt change.

A Sierra solutions engineer who shipped an enterprise customer-service deployment is a different hire than a Sierra research scientist who touched τ-bench. The former is useful. The latter is the one Noah Shinn calls back.

Who else is fishing the same 21 names

Instinct is not alone, which is why the sourcing window is measured in weeks, not quarters. In Refolk's index, the senior U.S. "AI agent" engineering pool across Member of Technical Staff, Research Engineer, Software Engineer, and Research Scientist titles is 77 people. Sierra alumni make up roughly 27% of it.

The named employers already hiring into the same cohort include:

• Replit, agent-first developer tooling.
• Zyphra, foundation model lab with consumer-agent surface area.
• Arcade.dev, tool-calling infrastructure for agents.
• Fundamental Research Labs, applied agent research.
• MAI Agents, pure-play agent startup.
• Meta, where Alexandr Wang's Muse just overtook ChatGPT as the #1 free app on both the Apple App Store and Google Play in the U.S. and Canada.

The Meta Muse hook is why this market broke open. Muse validated the consumer-agent category for every Series A board that was still unsure, and the expected downstream is 10 to 20 new agent startups fundraising on the Instinct template over the next two quarters. All of them will fish the same 21-person pool.

A 14-person raid is not a metaphor. It is the total addressable list, and most of them share a commute.

Describing that list in a Boolean string is where most sourcers stall. Current-employer:Sierra AND title:("research scientist" OR "member of technical staff") returns noise and misses former employees who have already moved. The gap Refolk closes: you describe the person in plain English and get a ranked shortlist with the τ-bench and Reflexion citations surfaced as evidence.

The counter-pitch that actually works

The one weakness in Sierra's grip on this pool is its in-person requirement. If you are hiring from New York, London, or remote, lead with location, not comp.

Sierra's engineer comp band runs $200K to $460K+ according to public estimates. Instinct is beating that on equity, not cash, by giving fractional ownership of a 14-person cap table. A third buyer, say a Series B consumer-agent startup from New York or Berlin, cannot win the equity contest and should not try. The lever is:

• Remote or hybrid flexibility for engineers who already quietly resent the SF-only mandate.
• A smaller, research-oriented team where evaluation infrastructure is the product, not a supporting function.
• A named technical co-founder or research lead the candidate wants to work with directly.

Shunyu Yao, Shinn's τ-bench co-author, is now Tencent's chief AI scientist. That is the prototype of what happens when you wait. Senior consumer-agent engineers move in weeks, through warm intros, and almost never through a job post. By the time a name hits a public hiring page, three competitors already have a signed offer in motion.

A 90-day plan for a founder who wants to clone this

The window to raid Sierra's engineering core closes roughly when Instinct's Series C cash finishes being deployed into recruiting, which is a Q1 2027 event. If you want a seat at this table, work backwards from that date.

1. Weeks 1-2: Map the 21. Pull the Sierra-tagged U.S. engineering list and tag by proximity to τ-bench and Reflexion. Discard the solutions engineers. You are left with 8 to 12 real targets.
2. Weeks 3-4: Add the 56 non-Sierra agent engineers. The senior consumer-agent pool is 77 nationally. Build the second tier from Replit, Zyphra, Arcade.dev, Fundamental Research Labs, and MAI Agents.
3. Weeks 5-8: Outreach with evaluation as the hook. Lead with the evaluation infrastructure you are building, not product strategy or market size. This cohort reads papers before pitch decks.
4. Weeks 9-12: Close on equity and research autonomy, not cash. You will not beat Sierra on cash and you will not beat Instinct on dilution. You can beat both on the quality of the research question and on the remote or hybrid structure.

What to watch next

The two signals that will tell you the raid is cresting are both measurable in Refolk's index week over week. One is Sierra engineering departures to companies with fewer than 50 employees. The other is new "AI agent" headlines appearing on profiles that previously said "research engineer" or "member of technical staff" at a frontier lab.

Expect both to spike through Q4 2026. The Meta Muse launch is the demand multiplier that keeps the wave coming. If you are a founder who wants to be inside the window, your pipeline needs to be built by mid-December, not mid-January.

The 14-person benchmark that Instinct set on September 28 is now the ceiling that every consumer-agent board uses to challenge "why do you need 40 engineers?" That math will not go away. Neither will the fact that the pool answering it is 21 people deep.

FAQ

How did Instinct really get to $10B with 14 employees?

The product is invite-only beta with annualized transaction volume above $1B, roughly 10% daily user growth, zero marketing spend, and 80% retention among the 40% of users who share a credit card within three weeks. Combined with Noah Shinn's authorship on Reflexion (91% HumanEval, NeurIPS) and τ-bench, those metrics made the round pricing a function of research pedigree plus unit economics, not headcount. Sequoia, Benchmark, and Coatue participated in the round on September 28, 2026, one month after a $250M Series B at $2.5B.

Is "Sierra alumni" really the right pedigree filter for consumer-agent hiring?

Only partially. Sierra's enterprise customer-service product is not the proximate skill, but Sierra's research-adjacent engineers (people who touched τ-bench, agent evaluation harnesses, or self-correction architectures) are. The 21-person U.S. engineering pool is the right starting list; the eight to twelve people inside it with evaluation-infrastructure work are the actual targets. A Sierra solutions engineer is a weaker signal than a non-Sierra engineer who shipped an agent evaluation framework in open source.

Why is Meta's Muse relevant to this hiring market?

Because it broke the "will consumers use agents" argument that every Series A board was still having. Muse overtook ChatGPT as the #1 free app on both the Apple App Store and Google Play in the U.S. and Canada under Alexandr Wang's push. The downstream effect is a wave of new agent-startup fundraises using Instinct's valuation as the comp, which means 10 to 20 new buyers chasing the same 21-person Sierra pool and the same 77-person senior consumer-agent pool over the next two quarters.

How fast does this pool move once a candidate is on the market?

Weeks, through warm intros, almost never through a public job post. Shinn's τ-bench co-author Shunyu Yao is already Tencent's chief AI scientist, which is the recurring pattern: by the time a name appears on a hiring page, three competitors have signed offers in motion. For this tier of candidate, a prepared pipeline (named list, warm paths, and a specific research pitch) beats an open requisition every time.