Agentic AI

There's a meaningful difference between an AI that answers and an agent that acts. Agentic AI is the latter — an autonomous agent that, given a goal, will take the steps to reach it: look up what it needs, remember what matters, decide what to do next, and know when to stop or call for help.

It's available as the Agentic AI action in your flows, and for many businesses it's the single most important block on the canvas.

What "agentic" means

A scripted bot follows a fixed path. An agent reasons. When a customer says "I ordered a blue jacket last week and it arrived damaged — can I exchange it?", the Agentic AI doesn't match a keyword. It works the problem: who is this customer? what did they order? what's our exchange policy? — gathering what it needs and acting, the way a capable rep would.

It does this by using tools — real capabilities it can call mid-conversation.

The tools it can use

The agent has a toolkit it draws on as needed. You don't script when each is used; the agent decides, based on the conversation.

• Know the customer — it can pull up context about who it's talking to: whether they're a returning visitor, what they've asked about before, where they are, and what pages they've been browsing. So it can personalize from the first reply.
• Search your knowledge — it can search your collections to answer from your real content, and filter precisely (size, price, stock) when the question calls for it.
• Remember and recall — it can save important details (a booking reference, a preference) and bring them back later, so the customer never repeats themselves within a conversation.
• Send messages — it replies in natural language, and can offer the customer buttons to choose from when that's clearer than free text.
• Resolve the conversation — when the issue is genuinely handled, it can close things out cleanly, with a short summary for your records.
• Hand off to a human — when it hits its limits or the customer asks for a person, it escalates with a written reason, passing the full context to your team. (See Live handoff.)

When you connect your store, it can also look up live orders and catalog data — which is how it answers "where's my order?" and "is this in stock?" with real, verified information.

Why one block can be a whole agent

Because the Agentic AI handles understanding, searching, remembering, resolving, and escalating on its own, a flow that's little more than Start → Agentic AI can run a complete support agent. You're not wiring every branch by hand — you're giving a capable agent a goal and the tools to reach it.

That's also why we suggest starting simple: get the Agentic AI working well with good knowledge and instructions behind it, then add structured flow steps only where you genuinely want tighter control.

Agentic AI vs. Ask AI

You'll see two AI actions in the toolbox, and the difference is worth knowing:

• Agentic AI — autonomous and multi-step. It can use tools, loop, remember, and decide. Reach for it when you want the agent to handle a request end to end.
• Ask AI — a single, focused turn from an assistant. Reach for it when you want one AI response at a specific point in a structured flow, without the full autonomous loop.

Both are documented in the AI & data actions reference. Most agents lean on Agentic AI for the heavy lifting and use Ask AI for precise, one-shot moments.