Agentic work is an AI workflow in which a model does more than answer questions: it plans steps, chooses tools, and helps carry out a task across multiple actions. In practice, this can mean drafting code, opening files, querying systems, creating tickets, or chaining prompts and tools toward a goal. The model may still need human approval, but it is no longer acting as a passive chatbot.
In cyber security, agentic work matters because added autonomy raises both capability and risk. Defenders use it to speed up triage, enrich alerts, and automate routine investigation steps. Attackers can use the same pattern to scale phishing, reconnaissance, malware development, or workflow automation. Security teams therefore treat agentic systems like privileged software: restrict tool access, isolate sensitive data, log actions, and require approval before any destructive or external-facing step. The key issue is not just what the model can say, but what it can do on behalf of the user.