CodeAct

The agent uses executable code rather than pure text as its primary medium for action and reasoning — it writes code, runs it in a sandbox, observes the typed result, and iterates, so a computation is executed rather… Typical conditions for use: Calculations, data transformations, or complex tool calls must be precisely executable; The result should be repro…

Use this profile to understand the building block briefly, place it in the model, and switch to the 360° assessment when needed.

Content type

Concept

Theoretical construct: explains a term, principle, or mental model.

Classification level

Structure

What organizes, connects, or makes decisions possible.

Available perspectives

Open this building block directly in an already curated view.

360°

In brief

CodeAct in brief

CodeAct lets an AI agent generate executable code as an action, run it and continue from the result.

Where does it come from?

Xingyao Wang and co-authors introduced the concept in their 2024 paper Executable Code Actions Elicit Better LLM Agents. It addresses the limits of rigid text or JSON action formats for complex agent tasks.

The mental model

The agent writes a small program as a work order, executes it, reads the result and adapts the next step.

Key building blocks

Core idea

Executable code provides a flexible shared action space for tool calls and control flow.

Practice context

An isolated runtime, permission boundaries and result checks are essential for safe use.

When is it useful?

CodeAct lets agents combine several tools and control steps in one executable action.

arXiv: Executable Code Actions Elicit Better LLM Agents

Why is this building block relevant?

The agent uses executable code rather than pure text as its primary medium for action and reasoning — it writes code, runs it in a sandbox, observes the typed result, and iterates, so a computation is executed rather…