Oh man, i feel u on this one. Honestly, I spent way too long sticking with GitHub Copilot Individual because it was the "standard," but it really started failing me once my Django projects got heavy on custom middleware and complex model logic. It basically felt like it was just guessing based on generic patterns instead of actually reading my project structure.

I also tried Tabnine Pro for a few months, but unfortunately, it was a bit of a letdown... the logic handling for deep unit tests was just not as good as expected and it kept hallucinating imports that didn't even exist in my repo. It was super frustrating having to fix the AI's "fixes" all the time.

For your situation, I would suggest checking out Cursor AI Code Editor. I switched a few months back and literally haven't looked back. Since it's a fork of VS Code, you dont lose your setup, but the way it indexes your entire codebase is just on another level. When I need it to generate pytest 8.0 suites for a Django view, it actually looks at my models and serializers to get the type hinting right. It makes Python 3.12 type hints feel way more natural because it actually "sees" the context of your whole workspace.

Ngl, it still trips up occasionally on really weird Django signals, but it's the closest thing to "professional-grade" I've found so far. Basically, if the tool doesn't have a deep index of your files, it's gonna give you mid-tier code. Good luck with the workflow upgrade!!

Ok so, after years of Django, my workflow journey ended up here:
- Cursor AI Code Editor: AMAZING logic, but $20/mo adds up.
- Amazon Q Developer Free Tier: Actually FREE and decent for tests!!

🙌

Just catching up on this thread... Similar situation here - I went through this last year when my Django projects started hitting that 'too big to handle' phase. I've been coding Python for over 10 years, so I'm pretty picky about logic accuracy. Honestly, I found that GitHub Copilot Individual—which was already mentioned—just couldn't keep up with my custom middleware or complex `prefetch_related` logic in larger repos.

I eventually shifted my workflow to Cursor AI Code Editor and the difference was actually wild. Since it indexes your whole codebase, the type hinting accuracy for custom Django Managers became way more reliable. For unit tests, I've been pairing it with Claude 3.5 Sonnet inside the editor. It handles `pytest` fixtures and complex mocking way better than anything else I've used over the years. It's the most 'pro' setup I've found for heavy Python work tho. gl!

Same here!

Tbh, if you’re looking for professional-grade Django support, you gotta move away from the big 'out of the box' services. Theyre okay for boilerplate, but for custom middleware and deep unit tests, you really wanna look into the DIY self-hosted direction. - Just go with a local inference server. You can't go wrong if you have the hardware to run it locally because the privacy and context handling are wayyy better for proprietary codebases.
- Look into open-source editor integrations. Basically, any tool that lets you point to your own local endpoint is gonna be superior because you can tune the system prompts specifically for Python and Django standards.
- Focus on a RAG-based setup. If you want accurate type hinting for a complex project, the tool needs to index your entire local repo. Setting up your own vector store for your code is pretty much the gold standard right now. It’s definitely more work to get running, but for a professional dev, the control you get over the logic handling is sooooo much better.

TL;DR: The discussion shows a clear split between mainstream cloud tools that struggle with deep Django context and more specialized indexing or local solutions. Honestly, having lived with AI in my Python workflow for a few years now, the long-term "ownership" experience is really about how the tool handles RAG (Retrieval-Augmented Generation). Most assistants start hallucinating when you have complex middleware or custom managers because they can't "see" the whole repo correctly. It becomes a question of how much time you want to spend managing the AI vs. actually coding. Just a couple of questions to narrow it down: what’s your actual budget for this? And are you in a position where you can use cloud-based indexing, or do you have strict security requirements for your proprietary code? Their are some very different paths to take depending on if your a solo dev or part of a bigger team.

Late to the party but I noticed no one mentioned looking at actual LLM benchmarks for coding performance. I mean, we can talk about vibes all day, but for professional Django work, you need the models that actually PASS the unit tests. Basically, I switched my workflow to Sourcegraph Cody recently because it lets you swap models on the fly. If you pair it with Claude 3.5 Sonnet, the logic for complex model relations is just SUPERIOR to what I was seeing with standard setups. It uses a very clever multi-file context strategy that actually understands how your Django apps talk to each other. Also, if you want something that actually writes the code for you based on those tests, you HAVE to look at Aider. It is a CLI tool that consistently tops the leaderboards for real-world editing tasks. It is way more reliable for generating those deep unit tests you mentioned because it literally runs the code and fixes it until the tests pass.

• Sourcegraph Cody (best for multi-model flexibility and context)
• Aider (best CLI tool for test-driven development and refactoring)
• Supermaven (if you need a massive 1M token context window for huge repos) TL;DR: Stop looking at just the IDE extensions and look at the underlying models. Cody with Sonnet 3.5 or Aider are the current performance kings for Python logic right now.