Bill Chen is an applied AI engineer and former founder of RiskAngle, a Y Combinator-backed startup that automated medical-record review for insurers and law firms. His work focuses on making coding agents dependable as models, APIs, and execution environments evolve.
Chen studied computer science at Columbia University, worked as a deployed engineer at Retool, and cofounded RiskAngle with William Hu. The company joined Y Combinator’s Summer 2024 batch and built software supporting insurance underwriting, claims processing, and legal case preparation. His parents’ work as life-insurance brokers and his experience filing a claim after a skiing accident helped shape the company’s focus. In a reflection on building RiskAngle, Chen emphasized customer needs, learning from rejection, and intrinsic motivation over investor attention.
Three-layer coding-agent architecture: Chen separates agents into their interface, underlying model, and agent harness: the prompts, tools, and execution loop that convert model capabilities into action.
The harness as an abstraction layer: Reusing maintained agent infrastructure helps startups avoid rebuilding integrations with every model release and concentrate on differentiated products and customer workflows.
Coding agents beyond software development: Terminal-based agents can organize files, analyze CSVs, and handle other computer tasks expressible through commands and tools.
Model-aligned integration: Chen distinguishes Codex models from the Codex agent and emphasizes matching tools and harness behavior to how a model was trained.
A coding agent depends on more than model intelligence: its tools, prompts, and runtime must fit the model’s habits without making every upgrade a rebuild.
