Issue 006: The Human Moat

One AI researcher was recently offered a reported $1.5 billion over six years to join Meta's superintelligence lab. He said no. He's not the only one turning down money that size. Eleven of eleven co-founders have now left xAI since its pivot into Elon Musk's wider empire. Meanwhile Anthropic, a company that openly refuses to match its rivals on salary, holds an 80% two-year retention rate that nobody else in the industry gets close to. Something has shifted in how the best technical talent in the world chooses where to work, and it stopped being the number on the offer letter.

This week is about what that shift means for anyone building culture inside a company where the product itself won't sit still. Five real examples of AI-era companies winning, and badly losing, this exact fight. Three frameworks for what actually replaces compensation once it stops being the deciding factor.

Katherine Berman is this week's Leader Spotlight. She's Head of People Experience at Snyk, a company whose entire product is making AI-generated code and AI agents trustworthy enough to build on. She joins The Work Life Reporter Live this Tuesday to talk about what trust looks like on the human side of a company that sells trust on the technical side.

Welcome to the sixth issue of The Work Life Reporter. This week you get:

Let’s get into it.

Anthropic's CEO Dario Amodei has said plainly that the company won't chase the top of the market on pay. Rivals have handed out signing bonuses reported in the tens of millions and multi-year packages worth over a billion dollars. Anthropic doesn't play that game. And yet independent tracking puts its two-year employee retention rate at roughly 80%, ahead of Google DeepMind's 78%, OpenAI's 67%, and Meta's 64%, in a market where a single senior researcher can be worth $600,000 to $1 million a year before anyone even discusses equity.

What Anthropic proves isn't that money stops mattering. It's that once every serious offer on the table is already life-changing, the marginal dollar stops deciding anything, and mission clarity starts doing the actual work. Employees there point to researcher autonomy, a flat structure with no forced management track, and a safety mission the company is willing to take real commercial hits for. That only works as a retention strategy if people believe leaving would cost them something a bank transfer can't replace.

xAI launched with a founding team built specifically around Elon Musk's stated mission to build a "maximally truth-seeking" AI, distinct from what its founders saw as safety-cautious rivals. Since xAI's merger deeper into Musk's wider corporate structure, more than 80 departures have been reported, and every one of the company's original eleven co-founders has now left.

The uncomfortable detail is that pay wasn't the story here. These were people who joined for a specific mission, at a company that could clearly still pay competitively. What changed was the mission itself, or at least their read on where it was heading once the merger reshaped priorities and reporting lines. Culture-first hiring cuts both ways: if you recruit people who joined for a reason bigger than the paycheck, a shift in that reason costs you the same people money never could have kept.

Mira Murati launched Thinking Machines Lab in February 2025 with around 30 people, roughly 20 of them poached from OpenAI, alongside senior researchers from Meta and Mistral. The company raised $2 billion at a $12 billion valuation within months. Less than a year later, five of its founding members had left for Meta, including one reportedly offered $1.5 billion over six years, three had returned to OpenAI, and one had joined xAI. The startup had lost the majority of the team it was built around.

The detail worth sitting with is what didn't fail here. Pedigree wasn't the problem. Capital wasn't the problem. Murati assembled arguably one of the most credentialed founding teams in the industry and backed it with more money than most startups ever see. What she hadn't built yet, by the time the novelty of the launch wore off, was whatever actually holds a team together once the headline fades and the work gets hard. A roster of stars is not the same thing as a culture, and the gap between them shows up fast when a bigger offer calls.

In July 2024, a faulty CrowdStrike software update crashed 8.5 million Windows machines worldwide, grounding flights and knocking hospitals and banks offline. Customer trust in the company dropped from 85% to 60% almost overnight. CEO George Kurtz took full public responsibility immediately, gave regular updates even while the full picture was still incomplete, and later introduced formal Customer Commitment Packages for those affected.

That's a customer-facing crisis story, but the mechanism behind it is exactly what the human side of trust requires inside a company too. What rebuilt confidence wasn't perfect information. It was leadership judgment about what people needed to hear and when, delivered consistently, including while the news was still bad. That's the same practice a people function needs during a layoff round, a strategic pivot, or any hard call that lands on a team, whether or not anything ever breaks in public.

Hugging Face pays its senior engineers an average total compensation of roughly $183,000, against $300,000 to $490,000 at Anthropic and $350,000 to $550,000 at OpenAI for comparable seniority. That gap runs into six figures, and Hugging Face still recruits and keeps researchers who explicitly choose it, at a company now valued at $4.5 billion, for open-source impact, real ownership over their work, and a genuinely remote-first culture rather than a hub-and-spoke office structure.

The mechanism here isn't researchers being naive about money. It's a trade the company names out loud rather than papering over. Hugging Face doesn't pretend the compensation gap doesn't exist or dress it up as something it isn't. It states the trade plainly: less cash, more mission and autonomy, and lets people choose with full information. That honesty is what makes the trade hold up over time instead of curdling into resentment the first time someone compares notes with a friend at a frontier lab.

AI compensation has split into two markets, not converged into one. Enterprise machine learning engineers earn $170,000 to $245,000 in total comp, while a small frontier-lab tier commands $600,000 to $1 million or more for the same job title. Equity now makes up 55% to 70% of total pay at the top of that market, up from 35% to 45% just two years ago. If your benchmarking deck is quietly measuring your offers against frontier-lab numbers, you're chasing a ceiling that was never built for your budget.

Most people leaders reading this aren't fighting the researcher-poaching war at all. That war is happening in a bracket most companies will never compete in, and trying to shadow-price against it wastes time and credibility with your own leadership. The actual fight for almost everyone else is retention within your real peer set, not the frontier tier making headlines.

Benchmark against companies you could plausibly lose someone to, not the ones making the news. If your comp conversations keep referencing frontier-lab numbers, that's a signal you're solving the wrong problem, not that you're underpaying.

Anthropic's 80% two-year retention rate against Google DeepMind's 78%, OpenAI's 67%, and Meta's 64% gets read as a hiring statistic. It isn't one. Those four companies can all pay senior researchers well into seven figures. The gap between 80% and 64% at that pay tier has nothing left to explain except what people actually experience once they're inside the building.

That reframes what a retention number is supposed to tell a people leader. Below a certain pay threshold, retention gaps are often a compensation story. Above it, a 16-point spread between companies that can all match each other on salary is a culture story with nowhere left to hide.

Stop treating your own retention rate as purely an HR dashboard metric. Once your pay is genuinely competitive within your market, a stubborn retention gap is telling you something about the actual experience of working there, not about the offer letter.

The old finding that happiness flattens out at $75,000 in income doesn't hold up. A 2023 reconciliation of the original research, run by the same scientists who disagreed on it, found no clean plateau at all: for most people, satisfaction keeps rising with income up to roughly $500,000. Only the unhappiest 15% of people stop feeling a benefit from more money, and even then, only around $100,000, not $75,000.

That matters because "money doesn't buy happiness past a point" has become a lazy excuse for underpaying people and calling it culture. It's not true for most people, most of the time. What is true is narrower and more useful: once someone is already being offered several life-changing amounts by several employers, as with the researchers turning down $1.5 billion, the differences between those offers stop being the thing that decides anything, because they've all already cleared whatever bar mattered financially.

Don't use flattened happiness research to justify a mission discount nobody's earned yet. Pay competitively first. Culture is what wins the close calls after that, not a replacement for the pay conversation.

Katherine Berman leads People Experience at Snyk, the company whose entire product is making AI-generated code and AI agents trustworthy enough to build on. She joined after several years at Toast, where she worked in workplace experience for the North America team.

Since then, she's set the direction for Snyk's People Experience function at a company reshaping how its own product changes the day-to-day work of the people who build it, described her team's purpose on LinkedIn in one line, build a workplace where the environment itself accelerates the mission, and taken on the role at the exact moment the AI talent market split in two, with elite researchers turning down record money for mission and everyone else watching pay converge fast.

In this week's Leader Spotlight, we sit down with Katherine to unpack what happens when a company that sells trust on the technical side has to build the same thing internally, why compensation is losing its power as the deciding factor for elite technical talent, and what any people leader can actually do about it, whatever their budget. A note on what follows: these are drawn from the themes Katherine has confirmed for Tuesday's conversation and her public positioning, not a transcript, since the conversation itself hasn't happened yet as this issue goes out. Come back after Tuesday for her actual words.

This is the language Katherine uses for her team's purpose at Snyk, and it's a useful test for any people function: does your environment make the mission easier to do, or does the mission happen despite it?

Katherine steps into this role at the exact moment the AI talent market splits in two: elite researchers turning down record-breaking money for mission and culture, while pay converges fast everywhere else. Expect Tuesday's conversation to dig into what actually decides who stays once a bigger offer stops being the differentiator.

Snyk's product is built to make AI-generated code and AI agents trustworthy enough to rely on. Katherine's argument, going into Tuesday, is that the people function is quietly solving the other half of the same problem inside the company itself.

Snyk's product changes what its own engineers' day-to-day work looks like, week to week. Katherine's angle for Tuesday is what stability and experience actually mean for people whose job is being reshaped in real time by the very thing they're building.

Not every people leader has a frontier-lab budget to compete with mission alone. Tuesday's conversation closes on what's actually available to any people leader trying to compete for talent, whatever their budget looks like next to a billion-dollar offer.

Thank you in advance, Katherine, we'll see you Tuesday!

Connect with Katherine on LinkedIn or find out more about Snyk at snyk.io

And that's a wrap for Issue 006. Something land? Hit reply and tell us which section you're taking back to your team.

Follow The Work Life Reporter on LinkedIn for what we share between issues, and join us live on Tuesday. Visit theworklifereporter.com for more.

PS. Know someone who leads people and would find this useful? Forward it on. Ten seconds to share, a lot longer to write.

PPS. Got a story, a leader, or a topic you'd like us to dig into? Reply with your ideas.