The First Move. The people and capital behind the AI race. |
How Small AI Labs Win Talent from the Giants, and How to Negotiate with Either Small labs can't outbid the giants on cash. They don't have to. Here's how the smaller labs are pulling people away from the biggest ones, and how to negotiate an offer from each side of the table. Written by Jack Leyden | Issue 04 |
| Together with Accelera Growth Partners, helping AI founders learn, build, and scale. |
In Issue 02, we followed researchers leaving the biggest labs to start their own. The next question is what happens after the founding team: how does a lab with fifty people hire away someone from a lab with five thousand, when it can't come close on pay?
The data says pay isn't what keeps people anyway. SignalFire tracked how many hires at each major lab were still there after two years, and the lab spending the most aggressively on talent kept the fewest.
Money doesn't keep people Share of employees still at the company two years after joining.   Source: SignalFire State of Talent Report, 2025. Covers hires through early 2023. |
Meta spends the most and keeps the fewest. That's the opening every small lab should be looking at. If people leave big labs for reasons other than money, a small lab can compete on those reasons.
A small lab can't win the auction. It can change what's being auctioned.
How to pull talent from the big labs, comp aside
The big labs have changed. Radical Ventures' head of talent argues they're now product-driven companies fighting for market share, and that neolabs emerged as a direct response. That's the gap small labs fill. Here's what they can offer that a big lab can't.
| |
| |
| |
| |
| |
|
Then match the terms that matter
You don't need to match big-lab pay, but you do need to match big-lab terms. The leaders have moved fast on vesting.
The disappearing vesting cliff How long a new hire waits before any equity vests.      Source: Wall Street Journal and HR Grapevine reporting on OpenAI policy changes, 2025. |
A candidate leaving a lab with no cliff won't accept a 12-month cliff at yours. Offer a short one or none for senior hires, spell out refresh grants, and put acceleration in writing if the company is sold or licensed away.
When a founder walks
Much of that pitch runs through the founders: the seat at the table, the direct access, the say in what gets built. That makes every founder departure a recruiting event. On July 29, Andi Peng worked her last day at humans&, the lab she co-founded with Eric Zelikman, Yuchen He, Georges Harik and Noah Goodman after working on reinforcement learning and post-training for Claude at Anthropic. Six months earlier, humans& had announced a $480 million seed at a reported $4.48 billion valuation. It had about 20 employees and no public product. Neither side has given a reason.
Founder exits at the neolabs The money stays. The founders don't always.
Sources: Andi Peng on X and RuntimeWire, July 30, 2026; TechCrunch; The Information and TechCrunch via Dealroom, September 2026. |
humans& isn't an outlier. Thinking Machines has lost four of its six co-founders and is reportedly raising at four times its last valuation.
For a small lab, a departure like this takes your best card off the table for every candidate who was joining to work with that person. Raise it before they do: who is staying, how founder equity vests, and what the research plan is without them. For a candidate, it belongs in the offer: if the founder you're joining for leaves, what happens to your scope, your title and your equity?
Five moves for small labs recruiting from big ones
Recruit the person, not the resume. Find out what they're frustrated by at the big lab. Process, scope and slow research are the usual answers, and all three are yours to fix.
Put the founders in the room early. Direct access is your strongest card. A first call with a co-founder beats three rounds with a recruiter.
Write the scope into the offer. "You'll own post-training" is worth more to the right person than another $50,000 of base.
Show them the plan for compute and runway. Big-lab candidates will ask. Have the numbers ready.
Plan for your own revolving door. Thinking Machines launched with six co-founders and has lost four. humans& lost one within six months of its seed. Small labs lose people too, and every departure is visible.
How to negotiate with a large lab and a small one
The same offer letter means very different things at a large lab and a small one. What's fixed at one is negotiable at the other. Here's where the room is on each side.
Where the room is What's usually fixed, and what's usually negotiable.
Mostly fixed Some room Wide open Ask about it
General patterns, not rules. Terms vary by lab, level and moment. |
Negotiating with a large lab
Base bands are tight, so don't spend your leverage there. Push on equity, sign-on and level. A strong competing offer is what opens the door: some large labs rarely offer sign-on bonuses without one, and those bonuses can come with repayment terms if you leave early. Ask how often tender offers happen and who can take part, because liquid equity is worth more than paper. OpenAI bought back about $7 billion of employee shares in August at its $852 billion valuation, according to Bloomberg. And if the lab is heading for a listing, as Anthropic reportedly is this fall, ask about the lockup and how your grant will be priced.
Negotiating with a small lab
Almost everything is open, so decide what you want before you start. Trade cash for ownership only if you believe in the company, and get the ownership as a percentage. Ask for the latest valuation, the runway and the plan for the next raise. Get your scope and title in writing. And ask directly what happens to your equity if the company is acquired or does a license-and-hire deal. Some teams in Issue 01 were paid well in those deals. Some weren't.
If you hold one of each
Use each offer to understand the other, not just to push the number up. The large lab pays more now and carries less risk. The small lab offers ownership, scope and a faster path to leading something. Decide which one you actually want before you negotiate. The best offers go to people who know why they're choosing.
Small labs won't win on money. They win on ownership, access and the work itself. Whichever side of the table you're on, negotiate for those.
If you got offers tomorrow, which would you take?
Get The First Move every week
Free, one issue a week on the strategy behind AI talent, capital and deals.
Know a friend who would want this? Forward it to them, or share thefirstmove.beehiiv.com/subscribe.
By the numbers
Who's joining the labs, what asking is worth, and how long people stay.
1. Joining the labs: this year vs last Engineers added at Anthropic, from an analysis of 1,680 public profiles.    OpenAI headcount growth by year.  Anthropic is hiring engineers faster this year than last. OpenAI grew 55% in 2025, but in January Sam Altman told staff the company would hire more slowly because it can do more with fewer people. Job postings still tell you where demand is: Anthropic opened 1,133 postings in the first quarter of 2026 alone, about half its total for all of 2025. Sources: independent analysis of 1,680 Anthropic engineer profiles (June 2026); Revelio Labs and makerstations workforce data; Financial Times. Lab headcounts vary widely by source because none of the labs publish them. |
2. What asking is worth For every $100 in a first offer, what tech candidates who countered ended up with.   
These are averages across tech, not lab-specific data. At the top labs, base bands are tight and the room is in equity, sign-on and level, so a similar percentage on a frontier-lab package is worth far more in dollars and moves through different levers. Sources: UCLA Anderson Review study of 3,858 tech candidates; Pew Research Center survey of 5,188 workers. |
3. How long people stay The labs are mostly new people.
About 9 in 10 people at OpenAI joined after late 2023  OpenAI had about 770 employees in November 2023 and roughly 7,850 by the end of 2025. Even with zero departures, nearly nine in ten people there today joined in the last two years or so. Short tenure at these labs mostly reflects how fast they've grown, not only how fast people leave. For how many stay, see the two-year retention chart at the top of this issue. Sources: independent analysis of 1,680 Anthropic engineer profiles (June 2026); OpenAI headcount from widely reported November 2023 figures and Revelio Labs. OpenAI share is derived from those headcounts and is a floor, not a measured tenure. |
Workforce figures are estimates from third-party data and vary by source.
Nothing here is legal, tax or investment advice.
