Claude Agent Enters IntelliJ IDEA Public Beta: Is Java Development About to Change?

Claude Agent joins IntelliJ IDEA public beta

On June 22, GitHub announced that Claude can now be used as an Agent Provider in GitHub Copilot for JetBrains IDEs. After installing the Claude Code CLI and configuring its path in IntelliJ IDEA settings, developers can start a new Agent session from the Copilot Chat Agent selector.

Why Java developers should care

Java projects often consist of multiple Maven modules and layers such as Controllers, Services, Domains, Repositories, Mappers, DTOs, configuration files, and tests. When an order interface fails, the call chain may look like:

OrderController → OrderApplicationService → OrderDomainService → OrderRepository → OrderMapper.xml → MySQL

Additional components like Redis, Feign, MQ, transaction interceptors, and permission checks may be involved, requiring developers to understand class references, bean injection, module dependencies, and runtime configurations.

From code completion to task execution

Traditional AI coding in Java mainly offers code completion—suggesting method bodies, DTO fields, or unit‑test snippets—without altering the development workflow. The new Agent mode lets developers hand over a relatively complete engineering task, for example:

请分析 order‑service 启动失败的原因。</code><code>先阅读当前启动日志、pom.xml、Spring 配置和最近的 Git 变更，定位失败原因后给出结论，不要立即修改代码。</code><code>如果确认需要修改，只允许修改 order‑service，避免改动 common‑core 和生产环境配置。

The Agent must read the exception stack, locate relevant beans, check dependencies and configurations, determine whether the issue is component scanning, circular dependency, version conflict, or missing environment variable, and then provide a verifiable conclusion.

Real Spring Boot failure handled by the Agent

Suppose a new risk‑validation feature causes the following startup error:

Parameter 1 of constructor in OrderService required a bean of type 'com.example.risk.RiskClient' that could not be found.

Traditionally, a developer would inspect the stack, search for RiskClient, verify Feign client annotations, configuration, and module dependencies, possibly digging into pom.xml, profiles, and conditional beans.

With the Agent, the developer can issue a richer task:

请只读分析当前 Spring Boot 启动失败问题，不要先修改代码。</code><code>错误提示为 RiskClient Bean 不存在，请按下面的顺序检查：</code><code>1. RiskClient 的声明方式和所在模块</code><code>2. order‑service 是否引入对应依赖</code><code>3. 启动类是否启用了正确的 Feign 扫描</code><code>4. 是否有 Profile 或 Conditional 注解阻止 Bean 创建</code><code>5. 最近 Git Diff 是否改变了相关配置</code><code>输出最可能的根因、相关代码位置、最小修改方案以及验证命令。

The Agent’s analysis can be corrected on the fly; if it goes down the wrong path, the developer can stop it and redirect the investigation.

IDE remains essential, now as a verification console

Even with powerful Agents, developers need a robust environment to review changes. Java errors often hide behind transaction boundaries, bean wiring, generic types, module dependencies, and runtime configs. IntelliJ IDEA will shift from “helping developers type faster” to “helping developers audit Agent work” through reference navigation, inspections, Maven dependency graphs, and debugging.

Team‑level agents and associated risks

Organizations can publish custom Agents that appear automatically in members’ JetBrains IDEs, enabling standardized code‑review, fault‑analysis, or upgrade checks. However, the current bypass‑permissions mode grants automatic approval for file edits and tool invocations, which poses risks for real projects.

Recommended safeguards:

Ensure a clean Git working tree before starting an Agent task ( git status).

Avoid running Agents in environments containing production credentials.

Execute complex tasks on an isolated branch or worktree.

After each task, review the diff, run tests, and build commands manually.

Getting started

Install the latest GitHub Copilot JetBrains plugin and the Claude Code CLI. Verify the CLI path with which claude (or where claude on Windows), then configure it under Settings → Tools → GitHub Copilot → Chat. Select Claude from the Agent picker.

Begin with read‑only tasks such as analyzing a startup exception, tracing an interface call chain, or generating test cases before allowing the Agent to modify code.

Three starter task categories

Fault analysis: Spring Boot startup failures, bean injection errors, Maven conflicts, test failures.

Impact analysis: Assess how changes to an enum, DTO, service, or database field affect other modules.

Test generation: Define a clear‑scope service, let the Agent list test scenarios, then generate JUnit 5 and Mockito tests.

Sample task template

任务背景：order‑service 在新增风险校验后启动失败。</code><code>当前目标：定位 RiskClient Bean 无法创建的根因，并给出最小修改方案。</code><code>分析范围：- order‑service - risk‑client 模块 - 启动类和 Maven 依赖 - 当前 Profile 配置 - 最近一次 Git Diff</code><code>禁止事项：- 不要修改 common‑core - 不要修改 application‑prod.yml - 不要执行数据库写操作 - 分析完成前不要修改代码</code><code>输出要求：- 根因判断 - 相关文件和代码位置 - 最小修改方案 - 可能影响范围 - 修改后验证命令

Conclusion

The Claude Agent public beta turns IntelliJ IDEA into an AI‑driven task controller for Java developers. While it can accelerate analysis and code changes, developers must still manage Git state, permission isolation, and diff review to ensure safety and correctness. The shift is not from writing code to abandoning it, but from manually performing every step to orchestrating the IDE‑based Agent to execute engineering tasks.