From our latest Radical Talks episode with Yvonne Likomanova, Senior Director of Talent at Radical Ventures
The most talented minds in AI are on the move: retention packages with nine figures, researchers boomeranging back, high-profile defections. Everyone’s tracking the exits. But the more consequential question isn’t where they are going and why.
Yvonne Likomanova, Senior Director of Talent at Radical Ventures joined Molly Welch to break down what she’s seeing on the ground — across frontier labs, hyperscalers, a new class of research institutions, and the applied AI startups trying to compete with all of them. The old playbook is broken, the market is more candidate-driven than ever, and the companies getting this right have figured out it isn’t really about comp anymore.
The Market Has Fractured
A few years ago, the path for top AI talent was relatively legible. You joined a frontier lab, a hyperscaler, or made a bet on an ambitious startup. Today that map has been replaced by something far more complex — Neolabs, AI infrastructure companies, neoclouds, applied AI startups, all competing for the same narrow pool of researchers and engineers.
The result is a market that is simultaneously more fragmented and more candidate-driven. But the more interesting shift isn’t structural, it’s psychological. Candidates are making shorter career bets, moving more fluidly between organizations, and operating with a real sense of urgency about staying close to the cutting edge. The belief that small teams can create outsized impact, faster than ever before, has meaningfully changed how people weigh risk.
The Neolab Phenomenon
The most significant talent story of the past year is the emergence of Neolabs. Few people have thought more carefully about what they mean for talent than Yvonne, who wrote about it in The Rise of Neolabs and the Talent Squeeze. Highly capitalized, research-driven AI companies that function more like private research institutions than traditional startups. Think Thinking Machines Lab, Safe Superintelligence, Pika Labs, and others — typically founded by high-profile researchers from frontier labs, pre-product, with extraordinary talent density and valuations to match.
The draw is straightforward: Neolabs are recreating what frontier labs used to feel like.
As the OpenAIs and DeepMinds of the world have scaled commercially, they’ve become something closer to large technology companies — more process, more bureaucracy, more pressure to ship. Researchers who joined those organizations to pursue foundational breakthroughs increasingly find themselves working within product-driven priorities. The launch of ChatGPT, for instance, fundamentally changed OpenAI’s internal DNA. Experimental teams were sidelined. The mission drifted toward deployment.
Neolabs offer the alternative: the freedom to pursue the research that actually matters to them, with peers operating at the same level, without the overhead of a scaled commercial organization. As Yvonne puts it, the distinction candidates are drawing is “if you want to invent the next paradigm, you join a Neolab. If you want to optimize the current one, you join a frontier lab.”
Whether this shift is permanent is less clear. Frontier labs still have unmatched compute, infrastructure, and scale. And Neolabs, if successful, will face the same pressures that transformed their predecessors. The more likely outcome is a cyclical dynamic — talent oscillating between institutions as the cultures and priorities of each evolve.
The Hyperscaler Boomerang
The hyperscalers — Google, Meta, Microsoft — are experiencing a version of the same dynamic. Talent is leaving for frontier labs, Neolabs, and startups. But the more telling story from the past year is the boomerang effect: researchers joining one organization only to return to a former employer within months.
Meta’s aggressive push to poach top researchers from OpenAI and DeepMind — at genuinely extraordinary compensation levels — ended with many of those same researchers being rehired at their original organizations, as priorities within Meta shifted and mission alignment came into question. Google, meanwhile, has leaned into the boomerang deliberately, with a meaningful share of its AI engineering hires being former employees recruited back.
The pattern reveals something important: compensation is no longer the dominant variable. The market isn’t purely comp-driven anymore. When candidates reach a certain level of financial security, they start optimizing heavily for autonomy, compute access, mission alignment, and the quality of the people around them.
What Engineers Are Actually Worth Now
AI is changing what the highest-leverage engineers do and therefore what’s valued. Implementation work, historically the bulk of a software engineer’s job, can increasingly be automated or compressed. What remains scarce, and increasingly prized, is product judgment and systems thinking: the ability to understand a business problem deeply and make the right architectural decisions without being handed a specification.
This has accelerated the rise of the product engineer and, more specifically, the forward deployed engineer (FDE), a role that’s become one of the most sought-after and least understood in applied AI.
FDEs sit at the boundary between the model and the customer. Deploying AI into enterprise workflows isn’t a one-time integration; it’s iterative, deeply contextual, and requires someone who can hold both the technical and the business reality simultaneously. The best FDEs are effectively running a small business inside a deployment — understanding customer requirements, adapting systems to real-world environments, and translating between what the research team built and what the customer actually needs. Former FDEs make disproportionately good founders, for exactly this reason.
The candidate profile varies. Some come from traditional CS backgrounds, others from machine learning research who want to be closer to the customer. But the supply is critically short regardless of path. For junior engineers navigating a less legible entry-level market, forward deployed roles represent one of the most compelling onramps available.
What Candidates Are Really Asking
The sophistication of candidates evaluating AI startups has increased substantially. Yvonne is now fielding questions from candidates that, a few years ago, were asked almost exclusively by investors: Does the company own proprietary data? Who controls the workflow? Are they training their own models, or are they a wrapper on top of foundation models? How would a frontier lab entering this space affect the company’s position?
The wrapper narrative has become central to how candidates evaluate applied AI opportunities. Companies without defensible moats: genuine data advantages, vertical integration, owned workflows, are facing real skepticism from exactly the candidates they most want to hire. And the perception of existential frontier lab risk, whether or not it’s warranted, can be enough to close off an otherwise compelling opportunity.
For founders, the implication is clear: you need to be able to answer the investor questions, because your best candidates are now asking them.
How to Compete
Two things, in Yvonne’s view, separate the companies winning on talent right now from those that aren’t.
The first is clarity of vision. The ability to articulate precisely what problem you’re solving, why it matters, what your unique insight is, and why you’re positioned to win — and then connect that directly to why this specific candidate’s work will shape those outcomes. It sounds obvious. In practice, very few early-stage companies do it well.
The second is intentionality about the early team. The first ten or fifteen hires are not just employees — they’re cultural co-founders. The values embedded in those early hiring decisions, the bar set for intellectual honesty and craft, become the signal that attracts or repels everything that follows. Early hires are also the most credible recruiting surface a company has before it develops a recognizable external brand. Getting them right is the highest-leverage thing a founder can control.
Looking Ahead
Yvonne’s view of the market two to three years out: increasing concentration of value around small groups of highly leveraged people. Teams stay leaner. Every employee — technical or not — expected to build and leverage agents within their own workflows. Traditional hiring signals like pedigree and brand names becoming less reliable as the people outperforming are often those who’ve continuously reinvented themselves to meet evolving needs.
The interviews will change too. More emphasis on learning velocity, judgment, adaptability, and curiosity. Less emphasis on credentials that may be poor proxies for the skills that actually matter now.
The companies that will win, in her opinion, are the ones that build systems where great people can continuously evolve and compound their impact over time.
This post is based on insights from Radical Talks, a podcast from Radical Ventures exploring innovation at the frontier of AI. For more conversations with leaders in AI, subscribe wherever you get your podcasts.