What is agentic AI?

Unlike standard AI tools that respond to prompts, agentic AI pursues multi-step goals autonomously — reshaping how companies automate work

By Ambia Staley2 min readUpdated July 17, 2026

Add QZ to Google

Agentic AI systems can autonomously break down complex goals, execute multi-step tasks using external tools, and adapt their approach without human direction at each stage.

Agentic AI refers to artificial intelligence systems designed to pursue goals across multiple steps, make decisions autonomously, and take actions in the world — rather than simply responding to a single prompt and stopping. While a conventional AI tool like a chatbot answers a question or generates a document when asked, an agentic system can receive a high-level objective, break it into subtasks, use software tools or external services to execute those tasks, evaluate the results, and adjust its approach — all without a human directing each step.

The core building block is the AI "agent," a software process that combines a large language model with the ability to call external tools: web browsers, databases, APIs, code interpreters, and other applications. A single agent can handle one domain. A more complex agentic system chains multiple agents together, each handling a specialized function, coordinated by an orchestrating layer. The result is a system capable of workflows that previously required sustained human attention — drafting and sending communications, querying and reconciling data across platforms, booking logistics, or monitoring and responding to changing conditions.

The business significance lies in scope. Earlier AI tools automated discrete tasks. Agentic systems can, in principle, automate entire processes end-to-end, compressing work that once took hours or days. That potential is reshaping demand across enterprise software categories: Identity and access management, for instance, faces new pressure because agents need credentials and permissions just as human workers do, multiplying the attack surface companies must secure.

The infrastructure demands are also distinct. Because agentic systems run longer, iterative workloads rather than single inference calls, their compute requirements scale differently from earlier AI deployments — placing new load on data centers and chip supply chains.

The central tradeoff agentic AI introduces is autonomy against oversight. The more a system acts independently, the harder it becomes to audit decisions, catch errors before they propagate, or assign accountability when something goes wrong. Managing that tradeoff — through logging, approval gates, and access controls — is where much of the practical engineering and regulatory attention is concentrated.

The essential business news, delivered fresh every morning.

Join 500,000+ readers who start their day with Quartz.

By subscribing, you agree to our Terms of Service and Privacy Policy.