Adam Tornhill argues code is key for agentic workflows

The author argues that code remains a crucial strategic medium in agentic workflows, serving as a precise mechanism for human-AI communication rather than just a human implementation artifact. By using inspectable end-to-end tests and code-based specifications, developers can maintain control, catch edge cases, and reliably guide AI coding agents. This approach combines high-level intent with rigorous automated verification to safely accelerate software delivery. Open article →