What is agentic AI?

The one-sentence definition

Agentic AI is software that can answer, reason, plan, use tools, and take action toward a goal — not just respond to a single prompt.

The five capabilities, in order

• Answer: respond to a question with a model.
• Reason: break a question into smaller steps.
• Plan: sequence those steps and decide what to do next.
• Use tools: call retrieval, code, APIs, databases, the warehouse.
• Act: change the world — send the email, refund the order, ship the deploy.

A chatbot covers the first capability. An agent covers all five. The jump is not the model — it is what the model is allowed to do, on whose behalf, and with what controls behind it.

Why agentic AI breaks differently

Traditional applications fail loudly: a stack trace, a 500, a timeout. Agents fail quietly. They take a plausible-looking action that turns out to be the wrong one. They loop. They drift. They make a confident decision on stale context. That is why the rest of the framework — Context, Control, Cost, Choice — matters more here than anywhere else in modern software.

Where agentic AI actually pays off

• Multi-step support workflows — diagnose, look up, draft, route, close.
• Internal operations — onboarding, reconciliations, refunds, returns.
• Data + analytics — answer in business language, not just SQL.
• Software engineering — read, modify, run, verify code.

What to read next

Start with Chapter 2 of the book, which walks through agents in plain English with examples your team will recognise. Then jump to Chapter 10 on why governance built for humans is not enough.