An agentic workload is an automated workflow in which one or more software agents repeatedly call a model, tools, or APIs to reach a goal. Instead of a single prompt and response, the system plans, acts, checks results, and retries as needed. Common examples include code assistants, research bots, ticket triage systems, and autonomous remediation playbooks.
In cyber security, agentic workloads matter because repeated model calls can amplify both benefits and risks. Defenders use them to scan logs, correlate alerts, enrich indicators, and trigger safe actions faster than manual review. Attackers can also abuse them to run large-scale phishing, recon, or malware-adjacent automation, especially when retries, branching, or tool use are weakly controlled. Poorly bounded agents may loop, overspend, or generate noisy actions that look like a denial-of-wallet event. Good controls include hard iteration limits, approval gates for sensitive actions, scoped credentials, rate limits, and monitoring for repeated failures or unusual call patterns.