The Tiny Teams Playbook

Shawn Wang surveyed top tiny teams at the World's Fair and the patterns are striking. Gamma serves 50 million users with 30 people through generalists, player-coaches, and culture of a small tribe. Gumloop aims to be a 10-person unicorn with extreme hiring standards including 4-day work trials and product-led hiring where top customers quit their jobs to join. Bolt.new hit $20M ARR in 60 days with 15 people by focusing on 10% of tasks that yield majority of results. Datalab hit 7-figure ARR with 7 people serving tier 1 AI labs. These aren't accidents - tiny teams are the multiplayer game of the decade of agents.

Hiring principles cut through the noise. Hire right or not at all - you have to be excited about the candidate or it's a no. Run paid work trials from 4 days to 3 months to be sure. Pay top of market salaries at 95th percentile and above. Build small crews under 15 people of senior generalists with far fewer juniors. Product-led hiring means your best customers become your team. Culture has to be low ego and high trust because trust equals speed equals ownership. Independence, grit, resilience. Radical transparency and accountability through wall of work and show and tells. Work closely with users, celebrate them, delight in feedback.

Operations kill the overhead. Almost no meetings means deep focus building instead of talking about building. AI Chief of Staff automates research and marketing through tools like Gumloop or Lindy. AI Support is well fleshed out now. Let fires burn in order to prioritize on the 10% that's critically important. Compound learning means Don't Learn It Twice - build reusable templates and playbooks. Either have an office or run very frequent AirBnB hack weeks because in-person matters.

Tech and product stay simple. Simple boring tech stack means shell scripts over Kubernetes, keep code modular. Simple product starts from UI wrapper over one API call to an LLM. Feature flags and experimentation are core principles. Create top tier internal evals for LLMs and harnesses, then market them. Every created benchmarks to eval models and market themselves. Evals are important but not enough teams create their own benchmarks and use them to improve product while building brand. Tiny teams compound speed by stripping everything down to what matters.