When we started building OpenTalent, we assumed the unit of hiring was the engineer. One profile, one set of preferences, one search. Everything we shipped in the first six months was built on that assumption. Then the pattern that broke that assumption started showing up — quietly, in support threads and in conversations with senior ICs — and once we noticed it, we couldn’t stop seeing it.
The pattern: senior engineers, when they move into a frontier lab, frequently bring one or two collaborators with them. Not in a “we’ll all apply separately and hope it works out” way. In a coordinated, intentional way, in which all parties know they’re moving together and have agreed in advance that none of them will sign without the others.
We did not build a feature for this for the first six months because we thought it was rare. We were wrong.
Once we sat down with the engineers who had moved this way and asked what made the difference between a group-move and a solo move, three things came up consistently. Not as theory — as the lived experience of senior people who’d recently done both kinds of move and could compare them directly.
// PART 1 · THE PATTERNWhat surprised us most.
The decision happens before the offer.
In a solo move, the decision to leave is usually made duringthe interview process — a candidate ramps from “open to looking” to “ready to sign” over the course of weeks, often nudged across the line by a strong final round. In a group move, the decision to leave has already been made before any of the group has interviewed. The conversation that matters happened in private between two or three people who trust each other, and the interview process is a logistics problem rather than a decision problem.
This compresses the entire timeline. Group moves close, on average, in 22 daysfrom first lab contact to all-signed. Solo moves at the same seniority average 51 days. The compression isn’t because the lab moved faster — it’s because the candidate moved faster.
The signal to the lab is fundamentally different.
Three senior people who have agreed to leave their current employer together are sending a signal that no individual candidate can send. It tells the receiving lab that the group has been functioning well enough that the members would rather keep working with each other than chase higher individual offers elsewhere. It also tells the lab that the integration risk is lower, because the group is already integrated.
Several hiring managers we spoke with described group offers using the same word — “anchor.” A group provides an anchor for a new sub-team in a way an individual senior hire doesn’t, because the new sub-team isn’t being constructed from strangers.
The economics tilt toward the group.
This was the surprise. We expected group moves to leave money on the table — that the joint-negotiation dynamic would compress comp to a kind of average. The opposite turned out to be true.The median per-engineer comp on group moves in our cohort came in 8–12% higher than comparable solo moves at the same seniority and lab, mostly through stronger signing packages and faster vesting cliffs.
The mechanism, when we asked, was straightforward: a lab paying for an anchor — a coherent piece of an existing team — is paying a small premium for de-risked integration, and the engineers capture that premium because they negotiated jointly and could credibly walk away as a group if the offer wasn’t competitive.
// PART 2 · WHAT WE BUILTThe feature.
What an engineer wanted, when they came to us with a group in mind, was a small set of things we hadn’t built. They needed a way to privately signal availability alongside one or two named collaborators, without those collaborators having to be active on the platform in the same week. They needed consent gates— if a lab expressed interest, no information about the group’s identities was shared until all members had approved. They needed joint-package handling: when an offer came back, it came back as a single coordinated package across the group, not as separate solo offers that the group had to back-channel and stitch together.
The feature is also explicitly nota way to do the thing people sometimes imagine when they hear “team move.” It is not a back-channel for poaching. It is not a way to coordinate raids against a current employer. The consent gates work in both directions: a group cannot be assembled around an engineer who hasn’t opted into a particular group. The model is closer to “small mutual-aid coalition between people who already trust each other” than “mercenary squad.”
What an engineer actually does in the feature is short. They create or accept a group invitation with one or two specific people. They set joint preferences for what the group is open to. They confirm that no information will be shared with a lab until everyone in the group consents. Then the group’s profile sits, quietly, alongside their solo profile, and they continue using the network as before.
// PART 3 · THREE STORIESWhat we saw work.
A 3-person post-training group from a US infra company.
Three engineers — one research lead, one infra owner, one eval owner — had been working together for fourteen months on a pre-training data pipeline. They liked the work. They didn’t like the company’s pivot away from frontier ambitions. They wanted to land somewhere that was still at the frontier, and they wanted to do it together.
They signaled in the network as a group in early Q3. By mid-Q4, they had a joint offer from a frontier lab that handed them a new sub-team focused on the same problem area, plus a signing structure that vested over two years rather than the lab’s standard four. They started in January.
A 2-person applied AI duo, transatlantic.
An applied research engineer in London and a product-leaning AI engineer in New York had collaborated on a customer-facing assistant for a year and a half. They were not on the same employer. They had become each other’s closest technical collaborator in spite of that. When the London engineer’s company restructured, they discussed it and agreed: if they were going to move, they wanted to land at the same lab, in the same office.
Their group profile had one constraint: same office, same team. Three labs declined to engage on those terms. One didn’t. They moved together to that one.
A 4-person eval team \u2014 the largest group we\u2019ve placed.
The largest group move we’ve seen was a 4-person eval team from a series-B AI lab that had quietly de-prioritized its eval investment. The team had built an internal eval harness that the new lab wanted; the new lab also recognized that the harness was the people, not the code. The offer was structured as the group acquiring a new team mandate, not joining an existing one — which let the lab pay an anchor premium and the group preserve the way they’d been working.
This is the group-move shape we see most often: a coherent eval, RLHF, or training team that the receiving lab wants to seed a new effort with, rather than slot into an existing structure.
// PART 4 · WHAT THIS TAUGHT USThe reframe.
We started this network on the assumption that the engineer was the unit. The data has been telling us, quietly, that the engineer is sometimes the unit and sometimes a fraction of the unit. A senior engineer in a high-trust working relationship with one or two others is a different shape than a senior engineer alone, and the hiring market treats them differently — even if the market hasn’t given that difference a name.
We don’t think this changes everything. The majority of moves in the network are still solo moves and probably always will be. Most engineers don’t have a tight collaborator they’d want to move with; most of the engineers who do, don’t happen to be ready to move at the same moment. The shape we’re describing is a minority of moves — maybe 6–8% of senior moves over the network’s lifetime — but it’s a minority with disproportionate outcomes for the people involved and disproportionate signal value for the labs hiring them.
The lesson we took, beyond the feature itself, is that career platforms tend to inherit the assumptions of the resumes they’re built around — one person, one history, one decision. Hiring at the senior end is increasingly not that. The teams that are doing the most interesting work at frontier labs were not assembled out of strangers. The hiring infrastructure for those teams hasn’t caught up to that fact. The feature is small. The implication is not.
// Read next
When quitting your frontier lab is the right call
Four subtle signs the senior ICs in our network watch for before pulling the trigger.
READ →// SCREENINGHow we read a frontier lab’s hiring signal
Five signals we use to figure out where a lab is actually hiring — beyond the careers page.
READ →// NETWORKTop 3%
What the network is, who’s in it, and how we vet the engineers who get matched to frontier labs.
READ →Thinking about a group move?
The feature is in the network — quiet by default, consent-gated in both directions. Join the network and start a group when the time is right.