Field Guide · 01, Foundations

What makes an AI agentic?

A chatbot answers. An agent acts, it takes a goal, decides the steps, reaches for tools, and keeps going until the job is done. Drag the core. Tap a bubble. Learn how autonomy actually works.

6 capabilities 1 loop 5 levels of autonomy ~7 min read
Agentdrag me
CAPABILITY · 6

The anatomy of an agent

Six capabilities turn a language model into something that can be trusted with a goal. Tap any bubble orbiting the core to see what each one really means, and why a plain chatbot has none of them.

Together they form a single loop: perceive the situation, reason about it, plan a move, act through a tool, observe the result, then do it again until the goal is met.

Tip, drag the glowing core and watch the orbit follow
02, The distinction

Generative gives you output.
Agentic gives you outcomes.

Generative AI is a brilliant one-shot machine: you ask, it produces. Agentic AI wraps that same intelligence in a loop and points it at a goal, so instead of handing you a draft, it does the work, checks itself, and comes back when it's finished.

Generative AI

The responder

Reacts to a single prompt
One turn in, one output out
Stateless, forgets after the answer
You drive every step
Produces content
" "Write me a restock email"
Agentic AI

The operator

Pursues a goal you set once
Many turns, until the goal is met
Remembers context across steps
It drives itself
Takes actions in the real world
" "Win back every buyer who lapsed this month"
03, The engine

The agent loop

Strip away the buzzwords and every agent runs the same five-beat cycle. It repeats this loop, sometimes hundreds of times, until the goal is reached. Hover a step to follow the cycle.

01
Perceive

Take in the goal and the current state, messages, data, the last tool's result, the world as it is right now.

02
Reason

Work out what matters, what's missing, and what the single best next move would be.

03
Plan

Choose a concrete action and the right tool to carry it out, the API, the query, the message.

04
Act

Execute. Call the function, run the code, send the reply, write the record. Reach into the world.

05
Observe

Read the outcome. Did it work? What changed, what's left? Then start the loop again.

Loops back to Perceive , autonomously, until the goal is met or it asks for help.
04, Under the hood

An agent is a model wearing a toolbelt

// the model coordinates context, tools, and review rules

The language model is the brain, but a brain alone can only think out loud. What turns it into an agent is the scaffolding around it: the parts that let it remember, decide, and reach into real systems.

1
Model

The reasoning core that interprets the goal and chooses each next move.

2
Tools

APIs, databases, search, code execution, the agent's hands in the real world.

3
Memory

Short-term scratchpad plus long-term recall, so context survives across steps.

4
Orchestrator

The loop controller that decides when to keep going, retry, or hand back to a human.

05, How far does it go

Five levels of autonomy

"Agentic" isn't a switch, it's a dial. Most products today live around Level 2; the frontier is pushing into Level 3 and beyond.

L0
AssistedYou do the work; the AI suggests as you go.
e.g. autocomplete
L1
ReactiveCompletes a single task each time you ask for it.
e.g. a chatbot
L2
OrchestratedChains tools to finish a defined workflow, you approve.
e.g. a copilot
L3
AutonomousOwns a goal end-to-end; escalates only on exceptions.
e.g. an agent
L4
Self-directedSets its own sub-goals, spawns sub-agents, refines its strategy.
e.g. agent swarms · emerging
Bring it to your business

Nantara is an agent,
not another chatbot.

It perceives your orders, reasons about your buyers, and acts on WhatsApp before you've finished your coffee. Everything in this guide, running on your store.