Corey J. Gallon is managing director of Rexmore, an AI-native holding company, and creator of chrome-agent, an open-source tool that gives coding agents direct access to working Chrome browsers. He builds systems that expand what autonomous software can do while keeping engineers responsible for architecture, validation, and understanding the finished product.
Earlier in his career, Gallon worked in PricewaterhouseCoopers’ finance analytics consulting practice and was a partner and head of finance at an institutional broker-dealer focused on algorithmic trading. He also taught data analytics, capital markets, and corporate finance at Loyola University Chicago, later became chief innovation officer of a consulting firm’s commercial technology business, and helped develop graduate machine-learning education. At Rexmore, he works across technology, investments, and education; he has been building AI coding agents since 2022.
- Human-directed agent development. Gallon’s AI Engineer Code 2025 framework assigns people ownership of architecture, interfaces, design tradeoffs, and acceptance criteria while agents implement specifications and tests. He treats generated software as an opportunity to deepen engineering knowledge, not an excuse to surrender it.
- Specification-driven implementation. A master project specification becomes small, individually documented features with explicit interfaces, validation contracts, and dependencies. Mapping those dependencies into a graph determines implementation order; agents receive focused context containing the relevant specification, completed dependencies, and execution guidance. Each finished feature becomes an atomic Git checkpoint.
- Multisensory agent feedback. Passing tests cannot reveal every broken interface or operational failure. Gallon combines rendered screens, application state, logs, errors, and real interactions so agents can determine both what failed and why.
- Browser-native automation. His chrome-agent exposes Chrome DevTools Protocol capabilities through a command-line interface, including browser state, network activity, console output, navigation, and input. His approach to real-world web automation uses a sense-act-verify loop, escalating from straightforward commands to more sophisticated interaction only when necessary.
