Skip to content

AI, decoded

What's the difference between agentic and non-agentic AI?

Non-agentic AI runs a fixed path: you give it an input, it returns an output, done — a chatbot answering a question, a model classifying a document. Agentic AI runs a loop: it sets a sub-goal, takes an action, observes the result, and decides what to do next, repeating until the task is done. The line is autonomy over the steps. Non-agentic systems follow a path you defined; agentic systems decide the path themselves, which is more capable and far harder to predict. Scaled Cognition's Dan Klein told Chain of Thought that a RAG question-answering loop isn't agentic to him; agentic means goals at play and actions that have consequences.

· Updated · Chain of Thought

Level 1: How AI works · 1.5 From chatbot to agent

AI Agents

  • 3 episodes
  • 3 guest clips
  • 8 glossary terms
A straight arrow with two fixed stops above a looping arrow, beside the question: fixed path, or a loop? Agentic vs. non-agentic AI.

In detail

Non-agentic: a fixed path

Much of the AI in production is non-agentic. Input goes in, output comes out, along a path the builder defined. A support chatbot, a summarizer, a classifier — these can be sophisticated, but they don’t choose their own steps. You can predict their shape because the control flow is fixed.

Agentic: a decision loop

An agentic system is built around a loop. It picks a sub-goal, takes an action — calls a tool, runs a query, writes to memory — observes what happened, and decides the next move. It keeps going until it judges the task complete. The model is no longer just generating text; it’s steering a process. A well-known example is the 2022 ReAct paper, which had a model write reasoning steps and task actions in alternation, so the reasoning could track and update the plan while the actions gathered information from outside sources. The ReAct entry covers the pattern.

Where workflows fit

The word “agentic” often gets used more loosely than this page uses it. Anthropic’s Building effective agents groups both kinds under “agentic systems” but separates workflows, where models and tools follow paths written in code, from agents, where the model directs its own process and tool use. By this page’s definition, a workflow is non-agentic even when it calls several models and tools, because its path is fixed in code, and that is what keeps it predictable. When a product is described as agentic, it’s worth asking which of the two it is. What makes an AI agent different from an LLM? covers the parts an agent needs around the model.

The trade-off

That autonomy is the whole appeal and the whole risk. Agentic systems handle open-ended, multi-step work that is hard to script as a fixed pipeline. They are also harder to predict, harder to test, and harder to trust, because the path can change from run to run. The question for any project is whether the task needs an agent’s autonomy, or whether a simpler fixed pipeline would be more reliable for the same job.

Agentic vs. non-agentic AI: a fixed path or a decision loop Left, non-agentic AI runs a fixed path the builder defined: input, step, a branch written as a rule (if this, go left, else go right), output. Examples are a support chatbot, a summarizer and a classifier; the shape is predictable because the control flow is fixed. Right, agentic AI runs a loop: pick a sub-goal, take an action such as calling a tool, running a query or writing to memory, observe the result, decide the next move, and repeat until it judges the task done. It handles open-ended, multi-step work, but the path can change from run to run, so it is harder to predict, test and trust. Bob van Luijt, episode 2: rather than if this, then go left, else go right, there's some reasoning to it. A fixed path, or a loop that picks its steps The line is autonomy over the steps: who decides what happens next, the builder or the system. NON-AGENTIC · A FIXED PATH input step if… left else right Every branch is a rule the builder wrote. Input in, output out, done. a support chatbot · a summarizer · a classifier Predictable: the control flow is fixed, so you know its shape. AGENTIC · A DECISION LOOP pick a sub-goal take an action observe decide next call a tool, run a query, write to memory done? stop Capable: the path can change run to run. The question: does the task need an agent’s autonomy, or would a fixed pipeline do the job more reliably? “rather than saying, if this, then go left, else go right, there’s some reasoning to it” Bob van Luijt (Weaviate), episode 2, on what gives a system agency. Autonomy also makes it harder to predict, test and trust.
The line is autonomy over the steps. The if-this-go-left framing is Bob van Luijt’s, from episode 2. Download the image

Hear it from the guest

“Rather than saying, if this, then go left, else go right, there's some reasoning to it”
Bob van Luijt, Weaviate EP 2 · around 4:59 · read the transcript
“If people are making security and safety decisions based on what you're seeing or what the video is seeing or whatever your AI model is doing … especially once you make it agentic … while now it's doing its reasoning on its own and it's choosing whether or not to take action.”
Kelly Vaughn, Spot AI EP 15 · around 16:43 · read the transcript
“But to me, that's not agentic. To me, agentic means there are going to be goals at play. There's going to be actions, and those actions are going to have consequences.”
Dan Klein, Scaled Cognition EP 54 · around 24:34 · read the transcript
Watch on YouTube

Quotes lightly edited to remove filler words.

From the conversation

This explainer is drawn from these episodes — each carries its full transcript.

Concepts in this explainer

Agentic AIAI CopilotGenerative AIReAct (Reason + Act)AI AgentLarge Language ModelRetrieval-Augmented Generation (RAG)Tool Use (Function Calling)

Keep exploring