Aaron Levie: Everyone is Wrong; We'll Have More Developers in 5 Years

More engineers in five years, not fewer

AI coding tools democratize engineering for the 85% of the economy outside Silicon Valley, enabling traditional companies like John Deere and Eli Lilly to hire technical talent previously accessible only to tech firms.

AI increases demand for lawyers

Generative AI creates exponential growth in legal content requiring human review, but institutional constraints (court filings, patent approvals, final sign-offs) remain the bottleneck, requiring more lawyers to handle the volume.

The rise of the Agent Operator

A new hybrid role (projected 500,000 to 1 million jobs) will embed technical experts in business teams to redesign workflows for agents rather than humans, managing MCPs, CLIs, and prompt syntax across fragmented enterprise systems.

Workflows must be redesigned for agents

Fortune 1000 companies cannot bolt AI onto existing human processes; they must fundamentally re-architect workflows around agent capabilities, requiring deep change management and data reorganization.

Automation exposes hidden bottlenecks

Streamlining patient referrals reveals the next constraint is still the 18-month wait for doctor appointments, proving automation shifts work to institutional capacity limits that require more human labor, not less.

Humans enter the loop at different points

AI doesn't remove humans but changes where they intervene, shifting from micro-task review to evaluating larger work products, which requires new apprenticeship models for junior professionals in law, banking, and medicine.

Value migrates from UI to APIs

As agents perform background work across systems, software value shifts from user interfaces and buttons to robust APIs and proprietary business logic (supply chain rules, accounting automation) that agents can programmatically access.

Agents increase API utilization

Software categories like ERP and CRM will see API call volume explode as agents continuously orchestrate work across databases, making the depth and security of business logic layers more valuable than the frontend experience.

Infrastructure requires constant maintenance

Enterprise AI implementations need continuous 'care and feeding' because new model releases break existing prompt syntax and agent workflows, creating ongoing demand for technical operators who bridge IT and business operations.