An agent wraps an LLM in a loop that reasons, calls tools, observes results, and repeats until done. The loop, not the model, is what makes it agentic.
Imagine asking a friend a hard question. A plain LLM call is like getting one shot to answer from memory: you reply once and that is it. An agent is like that same friend with a phone, a search engine, and a calculator. They can think, look something up, check the answer, look up something else, and keep going until they are confident. They are not smarter than before, they just have permission to take more than one step and use tools along the way. The repeating cycle of think, act, observe, decide is what makes a system an agent. Without that cycle, you just have a single answer from a single shot, no matter how powerful the model is.
Concept explanation~2 min read
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.
Concept explanation~2 min read
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.
The question looks elementary, but the answer matters because mislabelling a system as agentic or non-agentic leads to the wrong design choices. Calling a fixed RAG pipeline an agent invites people to expect adaptability it cannot provide. Calling a real agent a chain invites unbounded cost and latency to leak into systems that were supposed to be predictable.
The correct definition has one essential ingredient: iteration with tool use, where the model decides each next step based on the previous observation. Everything else, model size, training data, hardware, conversation history, is either a separate concern or a marketing distractor.
Why the loop is the defining feature
A plain LLM call is a function from prompt to completion. You give it text, it returns text, the call ends. There is no opportunity for the model to act on the world and react to the result.
An agent inverts that. The runtime presents the model with a state, the model emits an action, the runtime executes the action, the result becomes part of the next state, and the model is called again. The same model weights now participate in a sequential decision process. The behaviour of the system can change based on intermediate outcomes, which a single call cannot do.
This is why option B is correct. Reasoning, choosing a tool, acting, observing, and repeating until done is the minimal structure that turns a stateless inference call into a goal-directed loop. Remove any one of these moves and the system collapses back into a chain or a one-shot call.
Situations where this technique stops working.
2–4 min · Everything important, quickly.
Real products, models, and research that use this idea.
- Claude's tool use API lets a single model power both one-shot Q&A and full agent loops, the difference is whether the caller iterates on tool_use results.
- OpenAI's Assistants API exposes a thread with an explicit run loop, making the agent architecture visible in the SDK shape itself.
What an interviewer would ask next. Try answering before peeking at the approach.
QIf a chain calls a single tool and returns, is that an agent? Why or why not?
Walk through the four-move loop. A single call has reason and act but no second turn after observation, so the iteration property is missing. Lean on the path determination test: if the second step was decided in code rather than by the model, it is a chain.
Red flags & common mistakes
The phrases that signal junior thinking. Click to expand.
Red flags & common mistakes
The phrases that signal junior thinking. Click to expand.
Confusing model capability with system architecture. Bigger or fine-tuned models do not become agents, the loop with tools and observations is what turns any LLM into an agent.
60 second bullets to scan on the way to the call.
Define an agent as an LLM wrapped in an iterative loop with tool access.
Name the four loop moves: reason, act, observe, decide.
Primary sources. Browse if you want the original framing.
Same topic, related formats. Practice these next.