The New Startup Playbook | Box CEO Aaron Levie

AI could change the scale of a startup team: Aaron Levie says three to 10 people may be able to produce what once required 50 to 200. When Box began in 2005, its four founders were constrained by how quickly they could work at a keyboard; founders can now delegate more output to agents and focus on directing and reviewing it. That leverage is not automatic: agents make mistakes, need repeated attempts, and still require human review. Levie estimates advanced founders may gain three to 10 times the leverage from background agents, well beyond the 10–20% productivity gains previously reported for AI coding tools.

Coding agents are an early demonstration of the shift. Levie points to Replit, Cursor, Windsurf, Lovable, and Cognition, and says coding-agent companies could collectively reach hundreds of billions of dollars in market capitalization outside incumbent companies. As implementation becomes more commoditized, he expects strategy, clear specifications, market judgment, and product taste to matter more. Agents can help product engineers and product managers build directly when they know what they want, but a single all-purpose agent is not working well. Specialized agents for parts of a codebase or workflow may help address context rot, though the right architecture remains unsettled and depends on the team, company, and tools.

Small teams can also move faster because fewer people are coordinating their work with agents. Levie is learning from startups how to bring those methods into Box, but says large companies may adopt them more slowly: established workflows are deeply embedded, and engineering teams may lack the documentation, architecture, and tools needed to manage many background agents. The opportunity extends beyond software to fields including life sciences, healthcare, legal work, and consulting, where startups could compete with both traditional software companies and service providers.

For founders choosing an agent business, Levie recommends mapping industries against job functions and focusing on work they understand well enough to reverse-engineer. Having done the job being automated is the strongest fit; next best is having a co-founder with that experience, or learning the work through shadowing or direct experience. He sees many agent companies potentially reaching $1 billion to $10 billion or more in market capitalization, but cautions that some markets will be too small and many startups will fail to find adequate scale or market fit. Lower labor costs may expand demand: work offered at one-tenth the price might draw 50 times the demand. Levie contrasts a person costing roughly $100,000–$300,000, depending on the role, with an agent that might cost about $1,000. Cheaper agents could make services such as legal review accessible to more people and enable companies to take on functions they would not otherwise staff.

Levie expects much of the next wave of AI progress to benefit enterprise users, including pharmaceutical and life-sciences companies, because AI is fundamentally a productivity technology. Consumer opportunities remain, but current tools already meet many of his personal needs, including making songs, drawings, and stories for his children. By contrast, he says an AI model reviewing a law-firm contract still makes about 70 mistakes. Consumer startups may succeed through better or new app packaging, but he does not assume they will simply pivot from another business. Nor does he expect AI to fundamentally alter distribution: consumer products still need mechanisms such as virality, while enterprise companies still need a go-to-market engine.

A strong agent alone is not enough to build a durable company. Startups need access to a passionate audience through relationships and channels such as LinkedIn, X, and Reddit, supported by marketing, sales, and PR; without distribution, Levie warns, a company can be drowned out by 10,000 other startups and competing priorities. He says many core company-building principles remain timeless, but choosing co-founders matters even more when agents magnify the output of a small team. He also sees dropping out to start a company, once relatively rare among 20-year-olds in Silicon Valley, as a standard playbook for many young founders.