Chang Xu brings deep operator-founder experience from co-founding Onion Math, a China-based edtech company that raised around $90M, and serving as the first product manager at Minerva Project, giving her hands-on experience scaling venture-backed startups internationally before becoming a full-time investor. At Basis Set Ventures, she has developed a systematic thesis around infrastructure investing with four specific themes—"raise the ceiling," "lower the floor," "open space," and "data privacy"—and focuses on developer-led products where engineers drive adoption from the bottom up rather than through top-down sales.

• Identification of a massive shift in developer infrastructure needed to support AI agents. Argues that existing tools are poorly designed for autonomous workflows where agents act as explorers rather than script executors, specifically calling for new solutions in categories like browser automation and agent-native databases.
• Focus on the security and structural integrity of AI-generated code, often referred to as fixing the holes in vibe coding. Distinguishes between messy generation and structured engineering, highlighting the need for tools that ensure dependency trust and type-safety, such as BoundaryML for handling non-deterministic LLM outputs or Chainguard for supply chain security.
• Active experimentation with local development environments and the Model Context Protocol. details specific setups for creating a personal mixture-of-experts in the terminal, such as authenticating OpenAI and Gemini CLIs to work collaboratively inside Claude Code for solving hard architectural tradeoffs.
• Bridging the gap between academic breakthroughs and commercial application, specifically analyzing takeaways from NeurIPS. Points out distinct opportunities and failures in vertical AI, such as the lack of FDA approvals for generative AI in healthcare despite its predictive capabilities, or how multimodal fusion is solving the data-labeling bottleneck in robotics.