Agentic AI
AI systems that autonomously take sequences of actions — reading files, calling tools, making decisions — to complete longer tasks.
Agentic AI refers to AI systems that operate with a degree of autonomy, taking multiple sequential actions to accomplish a goal rather than responding to a single prompt. An agent can browse the web, read and write files, call APIs, run code, and make decisions about its next step based on what it finds.
In software development, agentic workflows look like: "Fix all the TypeScript errors in this project" — the AI reads the files, identifies errors, writes fixes, reruns the type checker, and iterates until clean. Claude Code, Cursor's Agent mode, and Devin are examples.
Agentic AI requires new thinking about oversight and verification. The more autonomously an AI acts, the more important it is to commit frequently, review diffs before accepting, and understand the scope of what you have authorised the agent to change.