Small Teams, Planet-Scale Intelligence

Think about what it took to build the telephone network. Thousands of engineers, decades of work, a corporate monopoly the size of a small government. Putting humans on the moon required 400,000 people across 20,000 companies and universities. Designing a new drug still takes, on average, over a decade and more than a billion dollars. The pattern held for centuries: if you wanted to do something that mattered at scale, you needed a massive organization to do it.

Software cracked that open a little. Instagram had 13 employees when Facebook bought it for a billion dollars in 2012 [1]. WhatsApp had 55 engineers serving 450 million users when it sold for $19 billion two years later [2]. A handful of people could build something a billion people used, as long as the thing they were building was software. Distribution and scaling got cheap. But the cognitive work - the design, the architecture, the debugging - still required human brains sitting in chairs, thinking.

AI breaks that constraint too. And this is the part that changes everything.

Cognition gets compressed

A study at GitHub found developers using Copilot completed tasks 55% faster [3]. That’s one tool, one task category, measured in 2023. Since then, AI has started compressing research, design, analysis, and strategy - not just coding. Each person becomes dramatically more productive, yes. But the thing that matters more is this: the minimum number of people you need to tackle a large problem is shrinking fast.

Consider what has already happened. DeepSeek, a Chinese AI lab with a fraction of the headcount and budget of OpenAI or Google, released a reasoning model in early 2025 that matched or exceeded GPT-4 on most benchmarks - reportedly for about $6 million in training cost [4]. Mistral AI, a French startup founded in 2023 with around 20 researchers, produced a 7B parameter model that outperformed Meta’s Llama 2 13B across most standard benchmarks [5]. These are not flukes or outliers. Open-source teams of five to ten people are building systems that compete with what the biggest corporate labs produce. A three-person team can now realistically train and deploy a competitive language model, build a global financial product, or run a research program across thousands of data sources. Five years ago that sentence would have been absurd.

The unit of production is changing

What this means is that the fundamental unit of economic production is shifting away from the large organization. It is moving toward small, highly capable groups of humans directing AI systems.

Picture what this actually looks like in practice. A few people set direction, make judgment calls, maintain taste and standards. The AI handles execution, search, synthesis, and scale. The humans decide what matters. The machines do the rest.

This has consequences that ripple in several directions at once. For startups, capital requirements for cognitive work drop fast, and the advantage tilts even harder toward speed and taste and contrarian thinking. Headcount stops being the bottleneck. Ideas and judgment become the bottleneck. For large organizations, it gets harder to justify competitive advantage through sheer scale when a small external team can match your output with AI amplification. And for governance, things get genuinely strange - when small groups can wield capabilities that used to be restricted to nation-states or major corporations, the regulatory and security landscape shifts in ways nobody has fully thought through yet.

What stays human

The scarce resource in this world is not computation. It is not even knowledge, which is becoming commodity-cheap to access and synthesize. What stays scarce is direction - the ability to figure out which problems are worth solving, to set constraints and values, to evaluate outputs with taste and judgment, and to make irreversible decisions under deep uncertainty.

These are fundamentally human capacities. They do not scale with parameter count or training data. They scale with experience, wisdom, and the kind of integrative thinking that comes from actually living in the world and caring about outcomes.

We are heading toward a world where small groups of very capable humans direct enormous AI systems to produce planet-scale impact. The question stops being “how many people do you need?” It becomes “how good is your team’s judgment?” That is the new constraint, and it changes everything about how we think about organizations, governance, and who holds power.

References

[1] Rusli, E. M. (2012). Facebook Buys Instagram for $1 Billion. The New York Times. https://archive.nytimes.com/dealbook.nytimes.com/2012/04/09/facebook-buys-instagram-for-1-billion/

[2] Olsen, P. (2014). WhatsApp: The Inside Story. Forbes. https://www.forbes.com/sites/parmyolson/2014/02/19/exclusive-inside-story-how-jan-koum-built-whatsapp/

[3] Peng, S. et al. (2023). The Impact of AI on Developer Productivity: Evidence from GitHub Copilot. arXiv:2302.06590. https://arxiv.org/abs/2302.06590

[4] DeepSeek-AI (2025). DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning. arXiv:2501.12948. https://arxiv.org/abs/2501.12948

[5] Jiang, A. Q. et al. (2023). Mistral 7B. arXiv:2310.06825. https://arxiv.org/abs/2310.06825