Andrew Ng: AI Hype Aside, Future Companies Will Be 10-Person Teams Rebuilding Data Architecture with Agents

Speaking at LangChain's Interrupt AI Agent Conference, Andrew Ng delivered a reality-check on the AI industry, arguing that while hype has exceeded expectations, the real transformation lies in how small, empowered teams use AI agents to rebuild enterprise data architecture. Ng observed that coding agent advancement has surprised even him. Six months ago he primarily used Claude Code; now he mixes OpenAI Codex, Gemini CLI, and OpenCode. Agents are evolving so rapidly that once-unthinkable workflows like writing production code on a phone are becoming natural. However, Ng highlighted that faster software development shifts bottlenecks. When code builds 10x to 100x faster, product management becomes the new constraint. Marketing, legal, design, and compliance all become friction points. A feature built in a day but needing a week for legal sign-off exposes hidden organizational drag. Ng is forming teams of one to ten engineers — generalists with high context who use AI not just for coding but for drafting marketing copy, terms of service, and product definitions. These small cross-functional units move at unprecedented speed with AI handling first drafts across multiple domains. On enterprise AI, Ng cautioned against treating AI solely as a cost-cutting tool. Cost savings have a ceiling; growth does not. He cited banking examples where AI enables "10-minute loan approval" by rethinking entire workflows. Call centers and drive-through ordering are other areas where AI drives growth through faster customer experiences. Ng's critical message centered on data architecture. Most enterprises have data in silos with permission systems designed for humans, not agents. For AI agents to transform businesses, companies must make unstructured data agent-ready. He predicted a wave of large-scale data restructuring as organizations realize agent readiness is the foundation for everything else. "The companies that benefit from agents," Ng noted, "will not be those that simply automate an existing process, but those capable of rethinking entire business systems around agent-driven workflows."