The author shares insights on using asynchronous coding agents like Claude Code and Codex to conduct code research. By framing research questions and assigning tasks to these coding agents, developers can leverage their capabilities to run experiments and provide results without constant supervision. This approach allows programmers to explore whether certain technologies, like Redis for app notifications, will work effectively by generating proof-of-concept code. Asynchronous coding agents operate on a "fire-and-forget" model, executing tasks on remote servers and filing pull requests upon completion. This method increases efficiency; the author mentions launching two to three code research projects daily with minimal time investment. He emphasizes the value of dedicated GitHub repositories for these tasks, arguing that separate repositories reduce constraints and risks associated with sensitive code, allowing agents to operate freely with full network access. The author showcases his public research repository on GitHub, which hosts various projects, including a performance benchmark of Python Markdown libraries and efforts to compile the cmarkgfm library for use in a Pyodide environment. He details specific prompts used to guide the coding agents, illustrating how they autonomously navigate tasks and produce results, even in complex scenarios that require compiling C extensions. This method of experimentation not only accelerates the research process but also fosters innovation in software development.