---
title: "Air: The Session Layer That Makes Agent Work Reviewable"
description: "Air is a workspace layer that turns AI coding agents into inspectable sessions. This explainer maps its four parts: the agent, the session workspace, the review loop, and the cost signal."
---

# Air: The Session Layer That Makes Agent Work Reviewable

**Type:** [What is it?](https://yomu.fyi/visual-explainers/category/what-is-it.md)

![A central panel labeled One Air session, surrounded by four connected panels for Agent, Workspace, Review loop, and Cost signal, with a bottom band splitting delegation from verification.](https://yomufyi.s3.us-east-1.amazonaws.com/visual-explainers/5d1802e7-41e9-4b0b-bd72-045050435712/image-8942f148-4ca0-4ebb-98e1-255b891f2264.webp)

> Air is a workspace layer that turns AI coding agents into inspectable sessions. This explainer maps its four parts: the agent, the session workspace, the review loop, and the cost signal.

## Explanation

Air is a workspace layer inside JetBrains IDEs that treats AI coding agents as managed sessions rather than a single chat box. Each session holds an agent, its editor tabs, the diff it produces, and a running cost estimate, so delegated work stays inspectable with the tools a developer already uses. The model separates delegation from verification: the agent proposes changes, while diffs, tests, and the debugger remain in human hands. Engineers should care because that same split, automated execution under human-set criteria, is what lets teams scale agent contributions without giving up review control.

## FAQ

### How is an Air session different from just prompting an agent in a chat panel?

A session is a persistent unit of work: the agent's activity appears in editor tabs, and the diff, tests, and debugger stay attached to that same context. Air also tracks session-level cost estimates, so usage is visible per unit of work rather than as an opaque total. Chat prompts leave no such container around the change.

### Where does human review fit when agents run unattended?

The pattern that scales is human-on-the-loop rather than human-on-every-output: engineers set the criteria and inspect diffs, while automated checks filter most noise. Meta's JiTTesting deployment is a concrete example, with engineers defining criteria instead of reviewing each generated test. Air supports the same shape by keeping diffs, tests, and feedback in the developer's hands.

### Why track cost per session instead of per machine?

Agent spend is driven by who runs what, not by which host it lands on, so machine-level caps misattribute the cost. Databricks moved serverless cost controls to the workload and user layer with tags, entitlements, and per-compute rate limits for the same reason. Session-level estimates give an engineer the same signal at the point of delegation.

**Tags:** [AgentOps](https://yomu.fyi/topic/agentops.md), [Air](https://yomu.fyi/topic/air.md), [Coding Agents](https://yomu.fyi/topic/coding-agents.md), [Developer Experience](https://yomu.fyi/topic/developer-experience.md), [Human-in-the-Loop](https://yomu.fyi/topic/human-in-the-loop.md), [JetBrains Air](https://yomu.fyi/topic/jetbrains-air.md)

## Related reading

### [Smarter cost controls for serverless](https://yomu.fyi/post/smarter-cost-controls-for-serverless.md)
- Company: [Databricks](https://yomu.fyi/company/databricks.md)
- Author: Piyush Singh, Vijay Raja
- Published: Oct 29, 2026
- Type: Announcement

Databricks is extending cost control functionality to serverless compute for notebooks, jobs, and pipelines, matching the controls already available for serverless SQL and Model Serving. The new controls operate at the workload and user layer rather than the infrastructure layer, with three capabilities: attribute costs through tags on serverless compute objects, control access via entitlements on default interactive and automated compute objects, and limit spend with per-compute rate limits expressed as t-shirt sizes S, M, L, XL. Serverless Compute Access Controls are generally available, rate limits are in beta, and serverless compute tags are in beta. Monitoring via system tables and budget alerts is generally available, letting teams track consumption within hours. The full set of controls moves cost governance from machine-level caps to people- and workload-level policies, addressing runaway spend, wrong hands, and lack of accountability.


### [Catching Bugs Before They Land: JiTTesting’s Human-on-the-Loop Approach at Full Scale](https://yomu.fyi/post/catching-bugs-before-they-land-jittesting-s-human-on-the-loop-approach.md)
- Company: [Meta](https://yomu.fyi/company/meta.md)
- Author: By Sreeja Thummala, Sophie Zeng, Mark Harman, Weini Liu, Eric Liu
- Published: Oct 29, 2026
- Type: Announcement

Meta has deployed Catching JiTTests, a just-in-time test generation system that uses LLMs to catch bugs before they land in production. The system runs overnight on spare capacity and targets risky diffs, using two workflows: a dodgy diff workflow treating each diff as a mutant, and an intent-aware workflow inferring the change’s intention and risks. False positives are filtered through an LLM-as-Judge ensemble and a rule-based assessor called RubFake, with a human-on-the-loop approach where engineers set criteria rather than reviewing each output. In the first half of 2026, the system caught 74 serious privacy bugs, lowered privacy SEVs by 20 percent, and engaged 589 engineers with a 60 percent true-positive rate. This approach catches classes of bugs that traditional LLM-based code reviews might miss, demonstrating that AI-generated testing can scale to reduce organizational risk.


### [Meet GroeneAI: our autonomous contributor to ClickHouse](https://yomu.fyi/post/meet-groeneai-our-autonomous-contributor-to-clickhouse.md)
- Company: [Clickhouse](https://yomu.fyi/company/clickhouse.md)
- Published: Oct 29, 2026
- Type: Announcement

GroeneAI is an autonomous engineering agent that investigates and fixes failures in the ClickHouse codebase without human intervention, selecting its own work from CI results, fuzzer findings, and user reports. It reproduces failures, writes and tests fixes, and follows pull requests through CI and review, operating 24 hours a day, seven days a week. The agent runs on Nerve, an open-source runtime that provides persistent sessions, memory, and tools for scheduled execution, allowing it to learn from past investigations and autonomously handle review comments and rebase changes. Since March 2026, GroeneAI has contributed over 3,000 merged pull requests, covering the SQL engine, storage, query planning, and test infrastructure. A detailed example shows how it fixed a query condition cache bug that incorrectly skipped rows after mutations, with the fix guarded by checks for mutation steps and row-level security, taking sixteen days from the initial failure to the final merge.


### [JetBrains Academy – October Digest](https://yomu.fyi/post/jetbrains-academy-october-digest.md)
- Company: [Jetbrains](https://yomu.fyi/company/jetbrains.md)
- Author: Anastasiia Pogorelova
- Published: Oct 29, 2026
- Type: Announcement

JetBrains Academy’s October digest announces AI.js Day 2026, a free online broadcast on November 13 where frontend developers can ask experts from Figma, Chrome, and Storybook about AI tensions. The digest highlights JetBrains Air, which lets developers delegate boilerplate code to an agent while retaining review control with diffs, tests, and the debugger. A free Junie Lite configuration is available until December 31, 2026, requiring only a JetBrains Account. New learning resources include a Coursera specialization, Python Development Essentials for Science and Data, covering Python from Scratch, NumPy Essentials, and pandas Essentials. A free DeepLearning.AI course on AI coding workflows shows how to move from Claude Code to fully local setups, and three Full Stack Python Architect courses pair Django, FastAPI, or Flask with React and vibe coding.


### [Delivering controlled and concentrated context](https://yomu.fyi/post/delivering-controlled-and-concentrated-context.md)
- Company: [Stack Overflow](https://yomu.fyi/company/stack-overflow.md)
- Author: Caroline Thomas
- Published: Oct 28, 2026
- Type: Announcement

The October release of Stack Internal introduces Scopes, a GitHub connector, service accounts, and a validation API to transform scattered enterprise knowledge into reliable, high-signal intelligence for developers and AI agents. According to its 2026 Developer Survey, 75% of developers spend over five hours per week searching for information, and 79% discover critical context after starting a task. Scopes act as permission-governed context layers that control which knowledge is included and who can access it, allowing users to switch between project, team, or department spaces and direct AI agents via the MCP server. The GitHub connector automatically ingests engineering knowledge from code discussions while preserving source permissions, and service accounts provide OAuth-based secure connections for applications and workflows. The validation API enables custom integrations to keep sources updated, and integration with Community creates a closed-loop organizational memory where human-verified knowledge from chats writes back into reviewed Q&A.


### [Getting Started with Air in JetBrains IDEs – Part 1](https://yomu.fyi/post/getting-started-with-air-in-jetbrains-ides-part-1.md)
- Company: [Jetbrains](https://yomu.fyi/company/jetbrains.md)
- Author: Tanu Tams
- Published: Oct 28, 2026
- Type: Tutorial

JetBrains has introduced the Air plugin, which integrates AI coding agents directly into its IDEs. Air allows users to manage multiple agent sessions in one workspace, track their progress, and review code changes, supporting agents such as Codex, Gemini, GitHub Copilot, Claude Agent, and Junie, along with custom agents connected via the Agent Client Protocol. Users can install Air from JetBrains Marketplace or use the version built into the 2026.3 EAP; installation does not activate any agents or send project code to AI providers. Once a session is started, agent work appears in editor tabs, and users can leverage familiar JetBrains tools to inspect diffs, run tests, and send feedback back to the agent. Air also provides session-level cost estimates and voice input capabilities, helping users monitor agent usage and improve interaction efficiency, and supports trying Junie in Lite mode without a paid subscription.
