PyCharm

With 90% of Django developers using AI in their weekly or daily workflows, the IDE is more important than ever for reviewing, understanding, and validating generated code. PyCharm gives you total control over your AI tooling alongside deep Django integration – without locking you into a single provider: 🤖 Bring your own agent: Run Codex, Claude Agent, Junie, or Gemini natively – or connect local models with Ollama and LM Studio. 💡 Teach your agent conventions: Encode reusable context using native skill support or the curated Skill Repository. 🐍 Django 6+ support: Complete template partials like ‘{% partialdef %}’ based on your project's specific Django version. 🔍 Review with confidence: Visual diffs and Local History let you inspect or restore code – even if changes were never committed. 🏗️ Visualize architecture and data: Navigate endpoints, inspect database changes, and view your project through Django Logical Structure. Read the full post on our blog to learn how PyCharm supports your AI-assisted Django workflow: https://jb.gg/p228oh #PyCharm #Python #Django #WebDevelopment #AI #JetBrains

With 263 fixes and improvements shipped across the PyCharm 2026.2 release line, we wanted to take a dedicated look at the updates that make the biggest difference in everyday Python development. Our team put together a detailed breakdown covering core improvements across the release: -SQLAlchemy 2.0 Integration: Improved type inference for `Session.get()`, seamless forward-reference resolution inside `Mapped[...]`, and support for `@name.inplace.setter`. -Smarter Auto-Imports: Prefers reusing existing module imports over creating redundant ones, with added auto-import support for nested classes. -Refined Type Inference: Better control-flow narrowing across loops, precise `Literal` types for enum `.value` and `.name`, and decorator-based parameter type deduction. -Clickable Type Inlays: View inferred type arguments directly in call sites and click them to jump straight to definitions. -Editor and Refactoring Upgrades: Active validation for f-string mini-languages and reliable reference updates when renaming modules. Read the full retrospective on our blog: https://jb.gg/2j0qju #PyCharm #Python #DeveloperTools #Programming #JetBrains

#marimo notebooks are available in PyCharm! 🎉 A new plugin developed by the marimo team brings reactive notebooks directly into the IDE. How are they different, you ask? ✅ Reactive cells and interactive UI ✅ Stored as Python files ✅ Git-friendly and runnable as scripts That means cleaner diffs, easier version control, and notebooks that fit naturally into existing Python projects. Install the marimo plugin now: https://jb.gg/pc-marimo-b

The agent knew what to do. It just picked the wrong Python. Wrong interpreter. Ignored virtual environment. Packages installed into system Python. PyCharm’s new Agent Environment Coordinator gives the agent your actual project setup – uv, Poetry, pip in a venv, or conda. Across 28 tasks and 6 models: 📈 Average task success: 68% → 98% 🧼 No polluted system Python The skill doesn’t take charge. It gives the agent accurate environment context and lets it decide what to do next. Ask your agent to install a package or run the project. It’ll pick the right Python automatically. 🔗 https://lnkd.in/eaRHfXcW 🔗

The biggest model isn’t always the best model for every part of an AI agent. In this new free course from DeepLearning.AI’s Andrew Ng and JetBrains’ Paul Everitt, you’ll learn how to use subagents, route tasks to different models, compare cost and performance, and move from cloud-based inference to a fully local setup with LM Studio and Gemma. Swipe through for the core ideas, then follow the full technical walkthrough and sign up for the free course here 👉 https://jb.gg/cxyw6k #AI #AIAgents #LocalAI #PyCharm #DeepLearningAI

I am excited about #PyConJP where I will give my #Japanese talk debut. どうぞよろしくね！Also there will be dedicated #PyLadies sessions. Come join us! #Python #AI #OSS #pyconjp2026 #パイソン #プログラミング #エンジニア

One of the most common #PyTorch beginner bugs isn't an error message. It doesn't crash anything, but your model trains worse than it should. The cause: forgetting one line – `optimizer.zero_grad()`. Here's why it matters: PyTorch adds up gradients by default. If you don't reset them in each batch, gradients from previous batches pile onto the new ones, and your weight updates get thrown off. The fix: Call `optimizer.zero_grad()` before `loss.backward()`, every iteration. Our tutorial walks through the full training loop 👇 Full tutorial + code: https://jb.gg/u67oqr

We know some of you rely on these plugins, so here’s what’s changing. In PyCharm 2026.2, we’re unbundling several low-usage plugins, including Data Wrangler, Hugging Face, Google Colab, and others. A more focused set means a leaner codebase and a faster IDE. Of course, the plugins continue to work and remain installable from JetBrains Marketplace. 👉 Learn more: https://lnkd.in/eBpxUBVS

Hand notebook work to an AI agent and it can quickly go wrong: ❌ Corrupted .ipynb files ❌ Lost models and data ❌ Tokens wasted while long jobs run PyCharm’s new Jupyter skill gives the agent a live kernel instead. State persists across cells. The notebook stays intact. Long-running jobs wait without constant polling. For Claude Opus 5, the kernel ran around 12% cheaper than the shell in our benchmark. Just open AI Chat and ask your agent to work in a notebook. Learn more: 🔗 https://lnkd.in/dQNmYD_V 🔗

Want an AI agent setup that’s faster, cheaper, and even runs locally? 🤖⚡ The next generation of AI development isn't just about picking the biggest model, it's about choosing the right model for each task. You’ll learn how to put this into practice in “AI Coding Workflows: From Cloud to Local”, a brand-new, free DeepLearning.AI course from Andrew Ng and JetBrains’ Paul Everitt. Built in PyCharm, the course shows you how to match the right model to each task, giving you control, choice, and managed costs. What you'll build This hands-on course walks through modern agent development using PyCharm AI Chat, showing how to optimize agents step by step: -SDD: Use spec-driven development to break big work into smaller phases for smaller models. -Subagents: Map this smaller work into tasks into specialized agents. -Smarter model routing: Use powerful models only when needed, while delegating simpler work to faster, cheaper models (the "big brain, little brain" approach). -Measure everything: Compare cost, latency, and quality as your architecture evolves. -Stay model-agnostic: Learn the underlying engineering principles, not just one specific tool. Explore today's open model ecosystem The course also demonstrates how to: -Configure OpenCode with OpenRouter. -Experiment with open-weight models like DeepSeek V4 Pro. -Apply the same optimization techniques across different models and providers. Run AI locally As local inference becomes increasingly practical, you'll also learn how to: -Set up LM Studio. -Run Gemma 4 12B locally. -Build a workflow that transitions from cloud-assisted inference to fully local agent execution. Whether you're exploring hybrid AI workflows or preparing for an entirely local future, you'll leave with practical techniques you can immediately apply. The course also uses a lightweight version of the Specification-Driven Development approach introduced previously in Paul and Andrew’s previous course. This is not a course about memorizing one model or framework. It is about learning reusable engineering ideas for a fast-changing AI ecosystem. Read more about the technical workflow and find out how to sign up for the course: https://jb.gg/cxyw6k #AIAgents #OpenWeightModels #LocalAI #AIEngineering #DeepLearningAI #PyCharm #JetBrains