20VC: Everyone is Wrong; We Will Have More Developers in Five Years | Why Frontier Labs Will Be Way More Valuable Than They Are Today | Are SaaS Companies Cooked: Which Thrive & Which Die with Aaron Levie, Founder at Box

What It Covers

Box CEO Aaron Levie argues that AI will create more engineers, lawyers, and knowledge workers over the next five years — not fewer — while explaining how enterprise agents require complete workflow redesign, new budget categories, and a new class of technical operator roles to function effectively inside large organizations.

Key Questions Answered

• •Developer demand beyond tech: The 85% of the economy outside the tech sector — manufacturers like Caterpillar, pharma companies like Eli Lilly, financial institutions — lacks sufficient engineering talent to automate their industries. AI coding tools like Claude Code and Codex now give these companies access to engineering capacity Silicon Valley has always had, expanding developer demand rather than contracting it.
• •Agent Operator as emerging role: A new role — roughly "agent operator" — will generate 500,000 to 1 million jobs. These workers need technical fluency in MCPs, CLIs, agents.md files, and prompt design, then apply that knowledge inside specific business functions like legal, marketing, or life sciences to redesign workflows around agents rather than human click-through processes.
• •Token budgets must move from IT to OPEX: Enterprise token spend cannot be managed inside existing IT budgets, which represent only 10–12% of revenue. Token allocation decisions will compete against marketing campaigns and operational expenditures instead, effectively doubling addressable technology spend and unlocking a budget category that has never previously been accessible to software vendors.
• •Data readiness blocks enterprise agent deployment: In a typical Fortune 500 company, contracts and documents are scattered across ten or more systems — legacy file shares, outdated document platforms, fragmented SaaS tools — making agent accuracy unreliable. Organizing data estates, connecting systems via clean APIs, and describing workflows explicitly to agents represents a decade-long implementation cycle for firms like Accenture in every major enterprise.
• •SaaS value shifts to API depth and business logic: Software platforms survive the agent transition based on how much proprietary business logic surrounds their APIs, not on user interface complexity. Platforms where agents consume data more frequently than humans ever did — unstructured content repositories, ERP systems with embedded supply chain logic — gain value; button-heavy tools with shallow APIs face structural pressure as agent interaction replaces human navigation.