Agent Client

An agent client is a program that actually turns on and runs an AI agent, letting users chat with it or hand it tasks. Code editors, chat apps, and terminal programs can all serve as agent clients.

Think of it like the relationship between a recipe and a microwave. If the tasks or abilities you ask an agent to do are the recipe, the agent client is the microwave that actually cooks according to that recipe. The catch is that every microwave brand has different button layouts and operating methods, so even the same recipe has to be reconfigured from scratch when you switch machines. Until now, the agent industry has worked much the same way—each company's agent client is different, making it hard to use a feature built for one directly on another.

Recently, moves to reduce this fragmentation have started to appear. Several companies are working together to create standards, aiming to let developers build a feature once and have it work identically across different agent clients.

In articles, it appears in phrases like "Build a plugin once and use it across compatible agent clients." Here, "agent client" is not the name of a specific product but a general term referring to tools that run agents, such as Cursor or GitHub Copilot. It shouldn't be mistaken for a specific company's product name.

Try entering the same task instructions or skill file into different agent clients (for example, an AI feature built into a code editor versus a terminal-based AI tool) and compare how the results or behavior differ. This makes it easy to feel what varies from client to client.