The Agent is The Loop

There’s a moment when a tool stops being a tool and becomes an agent.

For most people, that moment with AI happens in a chat interface. You ask, it answers. You ask again, it answers again.

But what if it could just… keep going?

Simon Willison recently added tool support to LLM 0.26, enabling models to execute functions and access external capabilities. This was the inspiration for taking it one step further.

The llm-loop-plugin does something deceptively simple: it gives Simon Willison’s incredible llm CLI the ability to loop. To keep working. To iterate until done.

Instead of:

• “Write me a Python function”
• Copy the output
• “Now test it”
• Copy that output
• “Now fix the bug”
• Repeat…

You get: llm loop "Create a Flask web app with a homepage and about page"

And it actually does it. File by file. Until it’s done.

This is the difference between a typewriter and a writer, between asking for directions and having a guide.

The magic isn’t in the AI model. The magic is in the loop. (well, truth be told, it is in the model)

The ability to persist, to iterate, to work toward a goal instead of waiting for the next prompt.

Most AI tools make you the bottleneck, constantly feeding them the next instruction.

The loop removes you from the critical path.

It lets the AI be what it was meant to be: agentic.

Install the llm-loop-plugin and give your LLM CLI superpowers:

llm install llm-loop-plugin

Then watch it work:

llm loop "Create a Flask web app with homepage and about page"

LLM is just waiting for you to close the loop.

# Join the Loop

The llm-loop project is open source on GitHub. Whether you want to report bugs, suggest features, or contribute code, your involvement helps make autonomous AI more accessible to everyone.