AI Coding Agents - From Copilot to Devin, Simply Explained

This article provides an overview of AI coding agents, which are tools powered by large language models that assist developers in understanding, generating, editing, and autonomously executing code.

Details

AI coding agents are a new class of tools powered by large language models (LLMs) that assist developers throughout the software development lifecycle. They range from inline autocomplete assistants that provide real-time code suggestions to fully autonomous agents that can independently plan, write, test, and deploy software. The article presents a diagram that categorizes AI coding agents into different types, including inline assistants, agentic CLI/terminal agents, AI-native IDEs, and autonomous agents. Examples of specific AI coding tools are provided, such as GitHub Copilot, Cursor, Claude Code, OpenAI Codex CLI, AWS Kiro, and Devin (Cognition). These agents vary in their level of autonomy, with inline assistants having lower autonomy and autonomous agents having higher autonomy. The article also discusses the importance of context awareness and the 'autonomy spectrum' in differentiating the capabilities of these tools. Factors like the agent's access to the codebase, understanding of the development context, and ability to plan and execute tasks independently determine where they fall on the autonomy spectrum.