Claude Code and What Comes Next

This article by Wharton professor Ethan Mollick demonstrates how advanced AI systems with proper 'agentic harnesses' can now perform sustained, autonomous work across multiple domains. The piece documents Mollick's experiment where he prompted Claude Code to develop a revenue-generating startup concept—the AI autonomously worked for 74 minutes, creating hundreds of code files and deploying a functional website without requiring intermediate human input.

The article explains the technical innovations enabling this capability: 'compacting' context to maintain continuity across extended sessions, skill-based architecture allowing dynamic tool selection, subagent delegation for specialized parallel processing, and Model Context Protocol integration for third-party tool access. Mollick argues this convergence of increased autonomous capability and sophisticated tool integration represents a significant inflection point, with similar harnesses for non-technical knowledge work imminent.