# Agent Development Environment

> An ambiguous term for a coding-agent workspace, an RL task environment, or an execution runtime. Specify which layer you mean.

- Source: https://www.zarifautomates.com/glossary/agent-development-environment
- Tags: ai-agents, reinforcement-learning, developer-tools

---

## 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](https://code.claude.com/docs/en/overview) is an example of this category.
- **Reinforcement-learning environment:** tasks, observations, permitted actions and scoring used to train or evaluate behavior. [Verifiers](https://github.com/PrimeIntellect-ai/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](https://modal.com/docs/guide/sandboxes), 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](/blog/best-ai-agent-development-environments) to choose between these layers, or browse the [RL environments directory](/blog/rl-environments-for-coding-agents) for concrete projects. For the broader concept, see [Agentic AI](/glossary/agentic-ai).