On a mission to make software development a more productive and enjoyable experience. Make it happen. With Code.
- Tweets 14,960
- Following 56
- Followers 215,206
- Likes 7,945
Pinned Tweet
Your IDE already knows your codebase. What happens when coding agents can use that knowledge too?
We tested it with Netflix using Claude Code and Codex. One finding stood out: the strongest models gained the most from IntelliJ IDEA’s code intelligence.
We connected IntelliJ IDEA’s code intelligence to Claude Code and Codex, then tested the setup with Netflix.
The first public results are out.
Typical efficiency gains: 20–30%.
Best code-navigation case: ~65%.
These results come from two real codebases. The efficiency metric averages token savings and time savings.
Correctness stayed stable or improved.
Here’s what we actually built.
The code engine behind IntelliJ IDEA ran headlessly, giving agents access to its capabilities through a CLI backed by a daemon.
LSP alone wasn’t sufficient.
Agents needed a convenient way to invoke the tools. The backend also needed to stay running and keep the project index warm between calls.
The idea is straightforward.
An IDE has already mapped the structure of your code and its dependencies — including what’s inside compiled libraries.
An agent can use that knowledge instead of spending time and tokens reconstructing it.
But the most interesting finding was which models benefited most.
The strongest models saw the biggest gains.
Especially at higher reasoning effort.
In these experiments, more capable models were also better at taking advantage of code intelligence.
My takeaway: better models and better tools aren’t competing bets.
You don’t have to choose between investing in model intelligence and giving agents better access to your codebase.
The two can reinforce each other.
Huge thanks to @pbakker and the Netflix team for working with us on this!
13
11
120
18,404
Junie Local now runs on @NVIDIARTXSpark from day one.
Get a full-fledged coding agent on your PC. It reads your code, edits files, and runs tests. There are no tokens or quotas, and no code leaves the machine.
Windows support is stable, too. All you need is a 24 GB Nvidia RTX card.
Try it for free: junie.jetbrains.com/local
ALT Image with a black background features text: "Your RTX Spark just got an agent. Junie Local runs on it from day one." The text is in white and green. The top corners display the logos for Junie and JetBrains. The bottom shows a green command line: "> /local".
6
14
78
7,815
Long-term AI adoption depends on driving down token costs. Mellum2.1 is our contribution to that shift – bringing an open-source 12B MoE model into local PyCharm workflows to deliver 140+ T/S without cloud overhead.
Learn more and see it in action:
Meet Mellum2.1 – JetBrains' open-source 12B MoE model built for coding agents and local execution.
🚀 140+ T/S in local testing scenarios
🇪🇺 Ethically trained European open model
💻 Efficient worker model inside IDE agent workflows
Try it locally: huggingface.co/collections/J…
5
9
83
8,531
Your IDE already knows your codebase. What happens when coding agents can use that knowledge too?
We tested it with Netflix using Claude Code and Codex. One finding stood out: the strongest models gained the most from IntelliJ IDEA’s code intelligence.
We connected IntelliJ IDEA’s code intelligence to Claude Code and Codex, then tested the setup with Netflix.
The first public results are out.
Typical efficiency gains: 20–30%.
Best code-navigation case: ~65%.
These results come from two real codebases. The efficiency metric averages token savings and time savings.
Correctness stayed stable or improved.
Here’s what we actually built.
The code engine behind IntelliJ IDEA ran headlessly, giving agents access to its capabilities through a CLI backed by a daemon.
LSP alone wasn’t sufficient.
Agents needed a convenient way to invoke the tools. The backend also needed to stay running and keep the project index warm between calls.
The idea is straightforward.
An IDE has already mapped the structure of your code and its dependencies — including what’s inside compiled libraries.
An agent can use that knowledge instead of spending time and tokens reconstructing it.
But the most interesting finding was which models benefited most.
The strongest models saw the biggest gains.
Especially at higher reasoning effort.
In these experiments, more capable models were also better at taking advantage of code intelligence.
My takeaway: better models and better tools aren’t competing bets.
You don’t have to choose between investing in model intelligence and giving agents better access to your codebase.
The two can reinforce each other.
Huge thanks to @pbakker and the Netflix team for working with us on this!
13
11
120
18,404
Read the case study 👇
blog.jetbrains.com/ai/2026/1…
6
2,411
We're proud to see @junie_ai, the JetBrains AI coding agent, featured as an official partner in the @figma MCP Catalog! 🎉
Another step toward bridging the gap between design and code, making developers' workflows smoother and more connected.
Excited to be part of this!
Announcing 20 new MCP partners at Config India (!!)
Amp, ClickUp, Dia, Junie by JetBrains, Linear, Meta Muse, Mistral, Notion, OpenCode, Sarvam Agents, Agentforce Vibes, Amazon Quick, Amplitude Wave, Cline, Delta, Dust, Glass, Gradial, Writer, Zed, Microsoft 365 Copilot
10
1
55
5,294
Try agentic coding with JetBrains Air and Junie Lite for free in PyCharm 2026.3 EAP 4 👇
Free, fast, and private agentic coding inside PyCharm?
JetBrains Air is live in PyCharm 2026.3 EAP 4, and you can test it for free with Junie Lite – simply sign in with your JetBrains Account. No paid AI subscription required!
🤖 Powered by open-weights Qwen 3.6 27B + JetBrains code intelligence
🔒 Private by design – your code is never used for training
⚡ Perfect for routine tasks, boilerplate, and test generation in Air.
1
3
54
7,253
A few AI power users don’t automatically make a whole team more productive.
At @WorldSummitAI, Ciara Byrne of JetBrains shared how teams can make AI agents useful beyond a handful of power users: build skills across the team, agree on shared practices, and automate repetitive work.
See how JetBrains Air can help your team: jb.gg/s9nrn4
18
3
27
5,116
Gemini 4 Argon is officially live!
@kristel_tech tasked two agents with building a web-swinging delivery game in a hyper-realistic night city:
• Gemini 4 Argon: $54.41 | 23 mins
• Opus 5.5: $56.97 | 2h 20 mins
Compare their performance yourself in JetBrains Air or Junie.
8
6
140
20,032
Companies are measuring lines of code again.
With flawed productivity metrics and spiraling AI spend, many leaders still don’t know what agentic development is actually costing them. Meri Williams (@Geek_Manager) calls metrics that look healthy while masking problems underneath “watermelon metrics”: green on the outside, red on the inside. That lack of visibility is one reason we made Air Governance, the organizational layer of JetBrains Air, our agentic development platform for engineering teams.
Get Meri’s full take on October 14, 4:00 pm UTC: jb.gg/lmrt3j
ALT Certainly! Here's a suitable alt text: "Webinar announcement by Blueprint. Title: 'Beyond Watermelon Metrics: The Actual Costs of Agentic AI and Paths to Productivity.' Featuring Ciara Byrne and Meri Williams.
3
1
28
5,436
Better yet, try it yourself.
Fastest way: download the 26.3 IDEs EAP and press ⌘2 (Ctrl+2), or look for the Air icon.
Using JetBrains at work? Air Teams and Air Governance have early access links in the thread. 👇
Holy moly, we’ve seen a lot of interest in JetBrains Air since the announcement!
-Air in IDEs: for developers
-Air Teams: for team leads & engineering managers
-Air Governance: for CTOs & engineering directors
Here’s how to try these things:
Links below 🧵
2
32
8,672
Financial institutions want agentic AI. Their reality is COBOL, legacy systems, and strict regulation.
Daniel Wolff and Aaron Wanjala discuss how to innovate without losing control – part of the wider industry shift reflected in @getsome_air.
Watch: jb.gg/8e9su0
ALT An online webinar banner titled "Adopting AI in Financial Services and Legacy Systems." The banner features two speakers. On the left is a person affiliated with Capgemini, and on the right is a person affiliated with Google Cloud Platform. The Blueprint logo is at the top left corner.
2
12
5,476
Air in JetBrains IDEs, in action.
Run your agent. Test the result. Verify the changes. All without leaving the IDE.
piped.video/yODck3N_lyU
3
4
51
16,069
Try it with Junie Lite for free.
plugins.jetbrains.com/plugin…
1
7
2,422
Live from @WorldSummitAI 👇
@kskrygan presents JetBrains Air at @WorldSummitAI
JetBrains Air is new system for building software with agents
1
3
34
6,156
"Just because you can build 100 things doesn't mean you should."
Engineering teams shared with Gleb Malkov, AI Solutions Architect at JetBrains, what they're learning about building with AI agents.
Dive into the 3 key challenges: jb.gg/midcid
ALT Two people kneel while working on a blue Formula 1 race car. Text above them reads, "3 things engineering teams tell us about working with AI agents.
ALT An illustration of hands holding a blue robot head. Below, text reads: "Automating the boring work creates a new operational burden.
ALT Two people interact near a blue Red Bull Racing vehicle. One is leaning toward the seated person, who appears to be in the vehicle. The text on the image reads, "Implementation got cheap. Review didn't.
ALT A person adjusts a blue, complex mechanical device against a plain backdrop. The image contains text that reads, “Check out the third most frequently mentioned scenario in our blog post.” The person is wearing bright blue pants.
3
3
22
6,630
This morning, our team breezed through the traffic and arrived at @WorldSummitAI the way many of us get around in Amsterdam: sure, it was a little windy, but that’s exactly how we like it!
Tomorrow at 3:30 pm, our CEO @kskrygan takes to the Marquee 1 stage with his keynote, “The AI Arms Race Was Only the Beginning.”
You’ll also find us at the JetBrains booth, where you can talk to the team and watch JetBrains Air demoed live.
See you there! 👋
4
1
12
7,189