What is an agentic loop?

It is the read-act-check-repeat cycle your program runs around an AI model. The program sends the conversation to the model, reads a field called stop_reason to see whether the model wants to use a tool, runs the tool and adds the result to the conversation if so, and repeats until stop_reason says the model is finished. That loop is what turns a chatbot that only talks into an agent that can actually look things up, run actions, and finish a multi-step task.

Does the AI model run the tools itself?

No. A tool is a normal function in your own program, described to the model with a name, a short description, and an input schema. The model can only ask your program to run a tool; your code decides whether and how to run it, then hands the result back. Anything you do not expose, the model cannot touch.

Why should the loop trust stop_reason instead of reading the model's words?

Because stop_reason is deterministic and the wording is not. A model can write text and ask for a tool in the same reply, or phrase "I'm done" a hundred ways. Every reply carries exactly one stop_reason: "tool_use" means run a tool and continue, "end_turn" means the work is finished. Loops that guess from the text stop early in ways that are painful to debug.

Why re-send the whole conversation every time?

Because the model has no memory between requests; each request starts from a blank slate. The growing transcript you re-send, including every tool result, is the only memory the model has. Leave a tool result out of the history and the model literally never sees it.

Agentic Loops: How an AI Actually Gets Things Done

What an agentic loop actually is: the read-act-check-repeat cycle that turns a chatbot into an agent which can look things up, run tools, and finish real tasks. A ground-up guide to stop_reason, tool calls, conversation history, and the one signal that tells your program when the work is done.

21 July 2026 Advanced

A complete beginner-friendly guide, no coding experience required. Every term is explained the first time it appears, every rule comes with the reason behind it, and one full worked example runs the whole cycle end to end.

Imagine a chef working from a recipe. They read a step ("chop the onions"), do it, look at the result, and then decide the next step. They don't read the whole recipe and magically produce a finished meal in one motion. They work in a loop: read → act → check → decide → repeat, until the dish is done.

An agentic loop is exactly this pattern, but for an AI. It's the machinery that turns a chatbot — something that only talks — into an agent — something that can actually do work: look things up, run calculations, send emails, update records, and string those actions together to finish a real task.

The whole point, in one line

A plain chatbot can describe how to check your order status. An agent, running an agentic loop, can actually go and check it and tell you the answer.

Sign in to view this build.

Create a free account to access the full build walkthrough on kevinsomany.com.

No account yet? Sign up free and come back here.

Agentic Loops: How an AI Actually Gets Things Done

Sign in to view this build.

Sign in or create an account.

Check your inbox.

Something went wrong.