Agent Development Environment
An ambiguous term for a coding-agent workspace, an RL task environment, or an execution runtime. Specify which layer you mean.
Definition
An agent development environment (ADE) is a workspace or task system used to build, execute, evaluate or improve an agent. The phrase does not name a single standardized product category. Specify the layer when using it.
Three common meanings
- Coding-agent product: an interface and tools that help a developer make changes to software. Claude Code is an example of this category.
- Reinforcement-learning environment: tasks, observations, permitted actions and scoring used to train or evaluate behavior. Verifiers provides infrastructure for this work.
- Execution runtime: the compute and isolation in which an agent’s tools run. A hosted sandbox, such as those described in Modal’s documentation, addresses execution rather than task quality.
An agent framework is another layer: it coordinates model calls, tools and state. It can use an environment or runtime without replacing it.
Example
A researcher defines a repository repair task with a clean starting revision and hidden acceptance checks. The agent edits files in a sandbox. The researcher scores the patch and resets the task for another run. The task is the evaluation environment; the sandbox is the runtime; the agent’s orchestration code is its harness.
Read the full ADE guide to choose between these layers, or browse the RL environments directory for concrete projects. For the broader concept, see Agentic AI.