Jargon:Agentic loop
Say it: ay-JEN-tik loop · 1 min read
Say this at the party
“An AI working in rounds: try something, check the result, adjust, and repeat until the goal is met.”
In plain English
Simon Willison has the neatest definition of an AI agent: a model that runs tools in a loop to reach a goal. The loop is the important bit.
Each round, the AI picks a step, does it, looks at what happened and decides what to do next. It's how you'd solve a puzzle: guess, check, try again.
The loop only works if the AI can tell when it's done. A test that passes or fails is perfect. "Make it better" gives it nothing to aim for.
Where it came from
The idea goes back to the 2022 ReAct paper, which had models take turns thinking a step and acting on it. In September 2025, Simon Willison argued that designing these loops is a key new skill, taking over from the art of the perfect prompt.
See it in action
You ask an agent to make a failing test pass. It reads the error, changes the code, runs the test, hits a new error, changes it again, and stops on the fourth try when everything passes.
What people get wrong
- A loop isn't a guarantee. Without a clear finish line, an agent can go round in circles and eat your allowance.
- A cap on tries, like "stop after ten", is good design, not a lack of trust.
Where Ugenie fits
A good loop needs a clear goal. Your three main tasks give Ugenie exactly that.
Get Ugenie →Go deeper
- ArticleSimon Willison, 2025Designing agentic loopsThe practical guide: clear goals, simple tools and a safe place to run.
- ArticleSimon Willison, 2025I think “agent” may finally have a widely enough agreed upon definition to be useful jargon nowWhere "tools in a loop to achieve a goal" comes from.
- PaperYao et al., 2022ReAct: Synergizing Reasoning and Acting in Language ModelsThe paper that started it: think a step, act a step, repeat.
Where it comes up
Related jargon
Drafted with AI, checked by the AI Council (Claude, ChatGPT and Gemini) and signed off by Phin.