🤖 The 2026.3 versions of JetBrains IDEs are out, with Air. Let agents do more of the work and stay in control of what ships.
- One place to review every change.
- Claude Agent, Codex, Copilot, Junie, and more.
- Your own subscriptions, or Junie Lite for free.
@ArtificialAnlys puts Gemini 4 Hydrogen at 51 on its Intelligence Index at the High thinking level, against 50 for the larger Gemini 4 Argon. Chemistry said hydrogen was the lighter one. Now in Junie: pick it from the model list.
Air is now in all 2026.3 JetBrains IDEs. We built it with your feedback, and we’re not done. @mhartington shares why we built Air, why we did so now, and what it means for you 👉 https://t.co/xHcgksVq26
I had the pleasure of talking at Apsara Conference and showing my white sneakers 🤣
We talked agentic AI: the future of sprints, guardrails for autonomous agents, and measuring real impact.
Watch the conversation 👇
The 10th JetBrains Developer Ecosystem Report is here.
AI agents are taking on more routine tasks, review overhead is a bottleneck, and code ownership is becoming a key question.
Read the full report: https://t.co/nqmToB8q6w
What stays in your workflow? What needs a patch in v11.0?
Thoughtworks Technology Radar Vol. 35 is OUT NOW! 🚀
Explore the latest tools, technologies and trends shaping software development.
Discover what to Adopt, Trial, Assess or Approach with Caution.
👉 Explore the latest edition: https://t.co/SzrpnwprQv
Nice DeepSWE score! Does it survive a lowercase filename?
Introducing MaxxBench: 89 DeepSWE tasks with small changes to the instructions or starting code. Same engineering problem, slightly different requirement.
Models clustered around 74% on DeepSWE end up 23 pp apart.
Good at coding, or good at DeepSWE? 🧵
we've completely rebuilt how AI agents work inside JetBrains IDEs
it's called Air, and starting with 2026.3 it ships with all our IDEs
Some features:
⛽︎ sign in with your existing ChatGPT, GitHub Copilot or Google account. Claude subscriptions work in terminal view
⛽︎ or bring your own OpenRouter, OpenAI, Anthropic or Gemini API key for supported chat agents
⛽︎ agents work right next to your code, so you can jump into any file they touch without leaving the IDE
⛽︎ hit a limit mid-task? switch to another subscription or API key right in the session, the work keeps going
⛽︎ Claude Agent, Codex, Junie and Copilot work out of the box. add Gemini CLI, Antigravity, OpenCode, Pi and more from the ACP registry, or connect your own agent through ACP
⛽︎ run agents in parallel in separate Git worktrees, then cherry-pick their changes back
⛽︎ agents get the IDE's symbol search, rename refactoring, inspections and run configurations, plus the debugger in IntelliJ IDEA and CLion
⛽︎ Junie Lite is free with a JetBrains account
⛽︎ leave notes on code or diffs and send them to your agent, Claude Code and Codex terminal sessions included
⛽︎ set up MCP servers once for every new agent chat, add skills from folders or GitHub
⛽︎ see at a glance which agents are running and which are waiting for you, use Caps Lock is LED of agents status
⛽︎ review the last turn, roll back individual files
⛽︎ stuck in a merge conflict? hand it to an agent right from the merge dialog
⛽︎ save agent presets and pin the prompts you keep reusing
⛽︎ macOS, Windows, Linux. Remote Development over SSH, WSL and dev containers too
and yes, you can change the chat text size
and much more features are coming before NY
Back in April I sat down with @iamitbeard (host of one of the largest Russian-speaking engineering podcasts). Almost 6 hours of non-stop talking later, I walked out feeling like I had completely overshared.
Today, the final 4-hour cut is live. While the interview itself is in Russian, the questions we unpacked are universal for anyone building software or leading engineering teams right now.
Instead of talking about AI hype, we went deep into the actual mechanics of Agentic SDLC, token economics, and how we build Junie at JetBrains.
Here is the breakdown of what we covered over those 4 hours:
My path from PHP developer and leading EPAM in the Netherlands to HFT, building our first GPT-3.5 prototype for traders, and co-building https://t.co/Jp8s3ZXSrV before joining JetBrains.
Where the real bottleneck in software engineering moved now that going from idea to prototype takes hours (and how JetBrains designers prototype directly with AI).
Spec-Driven Development: Where to store requirements and how to link specs, tests, and verification loops.
Why core engineering practices and tests matter more than ever when agents write the code.
Why JetBrains builds Junie when Claude Code and Codex already exist — and why we openly embrace 3rd-party agents in our IDEs via ACP ( plus an update on JetBrains Air and Junie since we recorded!).
If you speak Russian and want to watch links are below 👇
Big day for JetBrains.
Today we are releasing our 2026.3 IDEs with Air bundled in - a complete agentic workbench for professional devs.
Choose your agents, bring your subscriptions, and steer their work from your IDE.
Free, virtually unlimited agent Junie Lite.
More below 🧵
GitHub Copilot now works with JetBrains Air for coding tasks, guiding execution, and reviewing results.
• Run tasks locally in JetBrains IDE or in isolated cloud environments using your Copilot subscription.
https://t.co/MNT4ReTjZt
Junie Eco works with most LLMs, letting you get up to 2-3x more work done with the same token budget
Best value-for-money set up:
Junie Eco + GPT 6.1 Terra delivers roughly Opus 5.0 level quality
And $100 worth of tokens can offer even a heavy user about 3 MONTHS of use 🤯
How to keep focus and make decisions when we have to manage dozens of parallel agent sessions? Decision-First UX is the most important problem to solve in this multithreaded world of agentic development.
Running 10 coding agents in parallel means 10 chat streams to monitor, 10 context switches and states to reconstruct, and 10 walls of text competing for attention.
The bottleneck moved from writing code to human attention and judgment—decision-making. Those are the real scarce resources for builders today.
Even in an automated workflow (research → spec → code → review → fix), every human step left in the loop is a decision:
Right or wrong? Keep going or roll back? Option A or B? Safe to merge?
Yet tools still force a slow, chat-first loop:
Open tab → read transcript → reconstruct state → inspect diffs → decide.
To optimize for decision speed and quality, we need four shifts:
1. Portable task cards, not chat tabs
Decouple the task from the session. On laptop or phone, a card shows at a glance: goal, state, changes, test evidence, and the decision needed next.
2. An exception inbox, not status noise
Let repair loops run unattended. Interrupt only on ambiguity, broken checks, or risk.
3. Batched judgment, not live babysitting
As DHH noted on Lex Fridman's podcast, "the human in the loop is the limit." Let agents validate work asynchronously and clear a batched digest of ready decisions on your own cadence.
4. Step-specific decision surfaces
Assemble context for each gate:
• Spec: assumptions, options, Approve/Redirect.
• Code: test results + one-click test stand.
• Review: risk map before raw diffs.
The winning metric isn't how many agents you spawn—it's time to the next right human decision.
Which of these four is the biggest gap in your workflows today?
Alibaba just hosted Aspara - a big AI Conference in Amsterdam. Showcased Qwen models and their agentic platforms. For me personally huge moment was seeing the Head of Qwen Models showcase JetBrains and Junie on the main stage.
Kudos to the team of Junie Local 🙌
I participated in the panel and discussed why it is hard to move from cool AI demos to enterprises scale.
With 4–5 new models releases a week enterprises need a stable platform layer which gives the freedom to switch between models and agents while keeping workflow stable and predictable. This is exactly what we are building with Air Platform.
1/3 Not every coding step needs your most expensive model.
Meet Junie Eco Mode: your chosen model investigates and plans, then a lower-cost model handles implementation and testing. 🧵
~40 → ~70 tokens/s on M5 Max.
Bursts up to 107. All local.
Junie Local now runs on Splash + DFlash2, with lower memory use and support for M4 Macs (64 GB RAM).
https://t.co/wVWFUzlSUr