How talent became our most valuable signal

Why has talent become such a central force in the tech ecosystem? Answering that requires looking beyond how we measure talent to the incentives shaping the hiring market.

Game theory provides a useful lens to understand this. Hiring involves not only information asymmetry, which drives the need for job-seekers to convey information as discussed in our future of labor signaling article, but also the partial asymmetry of preferences, what is formally called a mixed-motive game (Kelly, 2003).

In economics, there are many examples of games where each side has a different motive, like a transaction in which the buyer and seller’s rewards have an explicit tradeoff. In the job market, although motives are partially aligned, each party has a different objective. Employers maximize productivity of their hires, whereas job-seekers are additionally interested in the reputation of the talent of their prospective coworkers, what Bargozzi and Cremer refer to in a 2024 paper as CfCA, or “concern for coworkers ability”.

While at a first glance, productivity and talent appear to be interchangeable, this is not quite the case. After all, if we wanted to strictly measure productivity in an economic sense, we could use some form of revenue/headcount. Talent is more complicated because it centers around a preference. Employees benefit from their coworker’s ability in two critical ways, neither of which is directly dependent on productivity:

1. They gain exposure to new technology, intelligent thought partners, and superior ways of working, which is especially important at AI labs carrying out original research.

2. They benefit from the reputation of their company’s talent in future career moves, whether it’s starting a new company or getting hired by a discerning competitor - think of the PayPal mafia for an obvious example.

So far, there’s nothing new or emergent in this picture. To grasp what’s changing in an increasingly talent-obsessed Silicon Valley, most notably among startups, we need to go one layer deeper, and understand “Endogenous Group Formation,” a lens developed in detail in a 2009 paper by Loury and Kim at Brown University.

Generally, endogenous group formation is when individuals form groups through voluntary participation. Loury and Kim specifically look at how subgroups form when heterogeneous populations have incentives to differentiate themselves in an externally visible way. This is like signalling, but instead of attaining "prestigious credentials,” the employees or founders create the prestige themselves through the groups they form.

When a group of employees from an existing company found a new startup, this is often a signal that this smaller group has a level of innovation, speed, or superior understanding that is not recognized or actionable in the larger entity. Similarly, when employees shift companies within the same field, it is often a move to increase the prestige of their resume to a higher degree than if they stayed. Of course, there are many reasons to move jobs, including the attainment of broad and varied exposure, or just boredom.

In today’s job landscape, however, this signal is increasingly telling.

Tech startups right now have taken the classic phenomenon of subgroup formation and accelerated it to the umpteenth degree. There is a zeitgeist around the AI industry which assumes that everything moves faster, but we can see this speed across all startups. According to 2026 data analyzing 4M+ job changes from Pin, tenure at early-stage startups is low, peaking at 16 months for Series A companies. As a benchmark, data released by the Bureau of Labor Statistics in late September shows the median job tenure across all jobs for ages 25-34 is 3.0 years and has remained steady since 2016. It’s safe to assume that tenure is ~half as long for an early-stage startup.

Frequent moves give us faster and more reliable signals on the relative abilities of tech startups to attract talent. These observations are all the more valuable given that a small number of talent-dense companies with proprietary, immensely valuable information and an unprecedentedly fast revenue ramp all exist in a few closely intertwined industries.

Talent moves fast, and quickly splinters from larger startups to new smaller startups, but there still needs to be an anchor in this calculation. In other words, it’s not just the interconnectedness of employee history that creates a level of talent, and talent is not just an ELO rating. If a top-tier employee leaves company A for company B, it could mean several things. On the one hand, company B could be superior to company A in career potential; on the other hand, given a very high reputation for talent at company A, company B may have expended resources in order to poach the employee. For these reasons, a functional algorithm needs objective features that are also informative of a company’s talent rating.

Fortunately, our proprietary data fills this need. The market is “seeded” with initial trade secrets on how AI works - for instance, there’s a strong argument that modern agentic AI is seeded by Google Brain’s (now DeepMind) development of the transformer and OpenAI’s development of GPT - participants are then free to move wherever they want.

The ripples of talent now, almost 10 years after the landmark “Attention Is All You Need” and 7 years after GPT-2, follow a similar pattern. Some of our top-ranked companies are formed from employees directly involved in the first AI revolution, but there are also newly forming patterns of academic discovery (either at a university or AI lab) and talent dispersion, some examples being the development of Denoising Diffusion Probabilistic Models for image/video generation in 2020, or how LLaMA (Meta AI, 2023) set off the wave of open-weight models.

As the use cases and adoption of AI multiply, there will inevitably be too many foundational discoveries to track individually, but Paraform’s Talent Density Index can inform efficient hiring for the companies driving technology forwards.