Return to site

🤖🛠️ JetBrains AIR: 🔟 INSIGHTS into AGENT-FIRST DEVELOPMENT with IntelliJ IDEA

What if AI agents could code, test, and debug inside your IDE?

November 21, 2026

🔸 TL;DR

In a recent episode of A Bootiful Podcast, Josh Long interviews JetBrains developer Anton Arhipov about how AI is transforming software development.

The conversation explores JetBrains Air, agent orchestration, spec-driven development, reusable skills, ACP/MCP integrations, and autonomous debugging.

Here are 🔟 insights from Anton's actual words, extracted from the episode transcript.

🔸 🔟 KEY INSIGHTS FROM ANTON ARHIPOV

1️⃣ [So the chat is in the center, in the middle.]

JetBrains Air experiments with an agent-first interface. Instead of writing code as the primary activity, developers describe tasks through chat while the IDE's traditional capabilities remain accessible.

#AgentFirst

2️⃣ [So we now have two surfaces inside the IDE.]

IntelliJ IDEA combines two workflows: the traditional code-centric editor and an agent-centric chat interface. Developers can switch between AI orchestration and manual coding without leaving their IDE.

#IntelliJIDEA

3️⃣ [I can, I can switch between the agents, whatever agent I like.]

Air provides a unified environment for multiple AI coding agents, including Codex, Junie, Claude, and others. Developers can select agents depending on their tasks, integrations, and subscriptions.

#MultiAgent

https://www.jetbrains.com/help/air/air-in-ides-overview.html

4️⃣ [It's, it's a huge session that was running overnight.]

Anton demonstrates an agent working for approximately six hours on a feature while he was away. The result required a few corrections but delivered substantial functionality. Air also tracks session costs.

#AutonomousCoding

5️⃣ [So there is a built in skills manager.]

IntelliJ helps manage reusable agent skills defined through Markdown instructions. Skills can be stored locally or inside Git-versioned projects to standardize development practices across sessions.

#AgentSkills

6️⃣ [Then the agent will be able to scan all those and generate the tests for me as well.]

Anton demonstrates a spec-driven workflow using features, use cases, acceptance criteria, and Markdown files. Agents interpret those specifications to generate tests and implement functionality.

#SpecDrivenDevelopment

7️⃣ [every new agent session doesn't have to scan the whole project]

Project rules, architectural conventions, and technology decisions can be extracted into reusable Markdown instructions, reducing repetitive exploration and potentially saving tokens across sessions.

#ContextEngineering

8️⃣ [it's integrating using the ACP.]

The Agent Client Protocol (ACP) standardizes communication between coding agents and IDEs. It helps JetBrains integrate different agents into a unified development and review workflow.

#ACP

9️⃣ [So there is an MCP endpoint in Intelli J]

IntelliJ's MCP server exposes selected IDE capabilities to AI agents, including project navigation, code analysis, and refactoring-related tools. Agents can leverage IDE intelligence instead of relying exclusively on text manipulation.

#MCP

🔟 [the agent actually identified the root cause and it fixed the test, right?]

Anton demonstrates an agent diagnosing a failing Java test with IntelliJ debugger integration. It uses logpoints to inspect execution, identifies the root cause, applies a fix, and verifies the result.

#AgenticDebugging

🔸 TAKEAWAYS

▪️ AGENT-FIRST DOES NOT MEAN IDE-LESS: The IDE remains essential for code understanding, navigation, review, and debugging.

▪️ SPECS + RULES + SKILLS: Structured instructions can make agent-driven development more consistent and repeatable.

▪️ ACP + MCP: Open protocols connect AI agents to development environments and their existing tools.

▪️ AUTONOMOUS DEVELOPMENT: Agents can implement features, execute tests, and investigate failures, but developers still need to review and validate their work.

▪️ HUMAN OVERSIGHT MATTERS: Delegating implementation does not eliminate engineering responsibility.

💬 Would you trust an AI agent to debug and fix your Java application autonomously?

#JetBrains #JetBrainsAir #IntelliJIDEA #Java #Spring #AgenticAI #AIAgents #SoftwareEngineering #DeveloperTools #CodingAgents #ModelContextProtocol #AI

Go further with Java certification:

Java👇

Spring👇

SpringBook👇

JavaFullstackBook👇