米汤
世界真小. It's a small world!
https://x.com/hiiinternet/status/2065117819948437765?
如果图显示不出来，可以看上面原文：
Builders, not researchers.
I pulled every LinkedIn profile that lists Anthropic as a current employer. 5,306 people. Kept the 1,680 who are actually engineers, then looked through 7,986 of their prior-role descriptions for what they did before they got there.
Here are the numbers.
They grew the org almost overnight.
?
15 engineers that are still at Anthropic were there before 2021. The org roughly tripled in 2025 (686 hires) and 2026 is on pace to match it (455 through June).
Half the current engineering org has been there under a year. 53% joined in the trailing 12 months. Median tenure: 10 months.
A giant org, built in about 18 months.
They almost exclusively hire senior engineers
?
Median experience before Anthropic: 12.2 years. Middle 50% runs 8.8 to 16.5 years.
Only 50 of the 1,680 have under three years of experience. 44% have 13 or more. New-grad hiring is basically nonexistent.
So the median hire has 12 years of experience and has been there 10 months.?
They index heavily on infra not really research
?
Infrastructure shows up in 40% of backgrounds. Backend, distributed systems, databases, and security each land around 20%. Reinforcement learning, the "RL" in RLHF, shows up in 3.3%.
The typical Anthropic engineer spent the last decade building large-scale production systems at a hyperscaler or an infra-heavy startup.
Self-listed skills say the same thing: Python 585, Java 566, C++ 443, JavaScript 376, SQL 302, Linux 230, Distributed Systems 189, AWS 154. The glamorous model-training work exists. It's just rare.
The #1 feeder isn't labs it's Google
?
Everyone assumes Anthropic raids OpenAI and DeepMind. Its biggest pipeline is Google, by a mile. The rival labs are the two small bars in the middle.
Anthropic over-pulls from places known for engineering rigor: Stripe, Databricks, Snowflake, Palantir, Airbnb.
Ever worked at, anywhere in their history: Google 405, Meta 273, Amazon 197, Microsoft 171, Stripe 124, Apple 87, Stanford 68, DeepMind 62, Airbnb 51, OpenAI 48. Half the org (50%) has FAANG on the resume somewhere.
They're also pulling from other labs OpenAI is a top-5 direct feeder, DeepMind top-6. Roughly 94 engineers moved straight from one frontier lab to them.
The PhD myth.
?
Only 13.7% hold a doctorate. One in seven.
The median hire is a senior engineer with a bachelor's or a master's, not a research scientist. The lab-full-of-PhDs image is mostly wrong at the engineering level.
Fields of study skew exactly how you'd expect a builder org to: Computer Science 819, then Mathematics 78, Physics 70, Computer Engineering 69. Philosophy cracks the top 20 (13) (safety?).
?
Stanford leads hiring heavily
?
Schools, all-time: Stanford 144, Berkeley 118, MIT 80, CMU 73, Harvard 42, Cambridge 39, UW 36, Waterloo and Cornell 35 each, Oxford 33, Princeton 32. Those top four are a quarter of the org.
80% of them share one job title.
"Member of Technical Staff."
A former Instagram CTO, ex-Adept founders, and Stanford faculty are all just "MoTS." The title is flattened on purpose. Seniority and function are invisible by design.
The one place "juniors" get in.
?
172 engineers have under six years of experience. 50 have under three. They are not generic new grads. They split into two archetypes, with almost no ordinary mid-level in between.
Look at how they differ from the org. More PhDs (19% vs 13.7%). Triple the rate of product/SWE titles (15% vs 5%). Far less likely to carry a FAANG resume (32% vs 50%).
What they have instead is pedigree that substitutes for years:
- The internship pipeline.?50% list internships at the following: Meta 16, Google 10, DeepMind 6, Microsoft 5, Amazon 5, plus Jane Street, Two Sigma, HRT, Optiver, Nvidia.
- Quant to lab.?9% came through elite trading shops (Jane Street, Two Sigma, Five Rings, HRT, Optiver, Citadel). Young math/CS competition types coming through HFT.
- Alignment fellowships.?6% touched MATS, SERI, Redwood, or ARC. A junior-only on-ramp that barely exists in the senior cohort.
The clean archetype: MIT, IOI silver medal, 2900+ on Codeforces, straight into RL and safety at four years in. They're screened on competition rank and publications instead of tenure.
They also skew more international than the seniors. Junior schools: Berkeley 15, Stanford 14, Cambridge 10, MIT 7, Tsinghua 7, Oxford 6, plus Imperial, NUS, Shanghai Jiao Tong, ETH Zürich.
So what do you do with this?
If you want to join Anthropic as an engineer, stop writing your resume for a research lab and write it for an infra company. Show systems you actually built and scaled. That's the resume getting hired. Early-career is the only exception, and there the bar is a top internship, a competition rank, or a paper.
If you're hiring against them, your target isn't a PhD or a lab logo. It's a senior builder from a hyperscaler or an infra-reputation shop, that's twelve years deep. Stripe, Databricks, Snowflake, Palantir. Anthropic is already fishing that pool hard.
---- :)
I'm recruiting engineers for startups backed by A16z, Sequoia, YC, Founders fund, Khosla, etc. Pre-seed to Series C, NYC and SF, Mid-Staff Product and AI engineers with salary ranges ranging from $150k - $400k
We'll tell you within a day if there are roles that fit your skills and interests.
Text our iMessage agent → 646-236-3745 and see who wants to talk to you (and how much they'd pay?you?not just the salary range) →?talent.fonzi.ai/home
