Building Ai Agents with Shell Scripts by Laurent Doguin

When considering an AI agent in its simplest form, it can be understood as a system with one or more inputs that takes action(s) based on those inputs, produces outputs for other actions, and continues this process iteratively. For developers, there are various approaches to building such systems, including workflow engines, no-code solutions, or simply coding. However, one of the most familiar concepts for devs is shell scripting, where CLI output is piped to other CLI tools using the "|" character. A powerful feature for one-liner, but devs can always do more thanks to flow control and variables supported in most shell. Over time, numerous shells have evolved, with new ones still being created from scratch,because devs still spend time in the terminal. For example, shells like zsh and fish have become increasingly popular, and Nushell is one of the latest additions, offering a native way to manage tabular data—ideal for handling structured inputs and outputs. This talk will explore AI agents from the perspective of a generalist developer, demonstrating how to build AI agents using Nushell and Couchbase as vector database.