UV for packaging and dependency management, Ruff for linting, Mypy for type checking (will be likely replaced by Ty when ready) and whatever editor people like (PyCharm, VSCode, Helix...)
What people use is going to be very environment biased. For example scientific research uses conda a lot, but I've rarely seen it used by code devs. People using only python will often have more system wide setup with tools and linters than those mixing different projects. Also you'll see some deprecated in tool use in projects which are older and not worth changing without a good reason. Etc.
With that out of the way: mise for managing versions of all the runtimes (asdf is fine too), uv for packages (previously poetry which is ok), mypy type checking for things I make. Otherwise I switch between projects a lot so just use whatever the projects use by default.
Deployment tools aren't really language specific (when they are they're rarely good). The editing, it's just Cursor, although I did like the jetbrains IDE previously.
But overall - just give things a go. It's usually quite easy to swap those tools as needed.
What I actually use at work is every version of Python from 3.8 on (though once we get out of crunch on some other things, I think I'll be able to sweep out everything up to at least 3.12), with venv/pip, and VSCode (with black for formatting and pylance for typechecking), and deployment is github->AWS CodePipeline->AWS Lambda for most things (and everything else is tools for the team where deployment is "clone the repo".)
If I had my choice, I'd switch environment and package management to uv.
Packages managed with pip and versions pinned with pipcompile. One ML production environment forces us to use conda we inherit from the pip environment and try to have as much done by pip as possible.
Formatted with black and flake8
Unit tests with unittest
Azure Dev Ops for running tests, security screening, and CI
The happy path for deployment is build docker images in AzureDevOps unfortunately we can’t do that for good reasons
I’m not a fan of typed Python, it doesn’t work as smoothly as you expect if you’re coming from a statically typed language.
Be careful enforcing pylint quality. It can be a great way to kill your teams velocity.
I’m describing a boring stack that works. All of these tools are 5+ years old and very well supported.
Not a python developer, just casually curious about the discourse. But I wanted to say that I completely agree with the point on letting dynamic languages be dynamic. Otherwise, you reduce the benefit of a dynamic language to the point where I feel like the value proposition comes too close to disappearing. Really at that point, you're just trading performance for compile time or lack there of.
Vim + PythonMode
`python -m venv` for local environment separation
Git + https://github.com/dandavison/delta for change management
When I need a debugger or a Git GUI, VScode, but Zed is becoming more and more interesting. These support LLM interaction, but things like Claude Code fit better with the above list.
Copilot works with various backend LLMs, including OpenAI, Anthropic, Google, and a few others I think.
We authenticate through a GitHub via SSO, which has a whole RBAC model behind it.
Maybe it would work with something else, but it’s usually more trouble than it’s worth, and depending on the source of the plugin for the other editors, could set off alarms with the security organization, and maybe legal as well. We have a lot of rules around our use of AI.
Can you use a dev container with a dockerfile and use UV for package management, ruff for formatting and lint, mypy for type hint checking, and pytest.
And put all of that on your CI as well so if they fail you don’t merge failing code.
I'm sorry, can't suggest much, I just use `Neovim` and `ripgrep` search.
No AI, no autocompletion, no fancy plugins.
At the moment I'm working for one of the biggest Python codebases: Odoo - a general purpose ERP from Belgium.
By that I meant everything related to python development starting from python version, package manager, environment manager, IDE and ending with deployment tools
By that I meant everything related to python development starting from python version, package manager, environment manager, IDE and ending with deployment tools
I think language servers and autocomplete are much more useful for .net (I used to do a lot of C# as a freelancer in college) than they are for python. Personally I just open a REPL and tab around in that if I need it but I do that maybe once a month. Most software is well organized such that you can keep enough of the theory in your head for the default nvim config to work easily. A small FITM model with a ring context like Qwen can help a bit with boilerplate if you really want autocomplete.
- uv for environments, packaging, and dependencies.
- ruff for linting and all the other stuff it does.
- mypy for type hint checking
- VS Code (although I'm playing with zed.dev too)
UV for packaging and dependency management, Ruff for linting, Mypy for type checking (will be likely replaced by Ty when ready) and whatever editor people like (PyCharm, VSCode, Helix...)
Not a big company, however:
- uv (went from pipenv to poetry to uv), which was the right decision imo.
- ruff (replaced black)
- PyCharm, but some people use vscode or neovim.
- mypy for type checking, which I am not 100% happy with. One experienced developer swears by mypy, thats the reason we haven't explored alternatives.
- 'just' as task runner
uv is just such a fast and reliable package manager. It implements the "lock" problem python has, with the uv.lock
UV has become my goto for package/venv management. It can work as a close to dropin replacement for pip.
Ruff is my goto for linting and formatting. replacing black, isort, flake8, etc.
There's lsps for both, so you can use them in vscode or wherever you're doing your editing.
At EntwicklerHeld we used a setup i was happy with:
- poetry for dependency management
- black for formatting
- flake for linting (you can use ruff now)
- PyCharm as IDE
- pyenv for python version management -> but I switched to devbox a year ago and never go back
- docker as end artifact
- deployed to kubernetes (at the end k3s) with ansible and terraform
What people use is going to be very environment biased. For example scientific research uses conda a lot, but I've rarely seen it used by code devs. People using only python will often have more system wide setup with tools and linters than those mixing different projects. Also you'll see some deprecated in tool use in projects which are older and not worth changing without a good reason. Etc.
With that out of the way: mise for managing versions of all the runtimes (asdf is fine too), uv for packages (previously poetry which is ok), mypy type checking for things I make. Otherwise I switch between projects a lot so just use whatever the projects use by default.
Deployment tools aren't really language specific (when they are they're rarely good). The editing, it's just Cursor, although I did like the jetbrains IDE previously.
But overall - just give things a go. It's usually quite easy to swap those tools as needed.
What I actually use at work is every version of Python from 3.8 on (though once we get out of crunch on some other things, I think I'll be able to sweep out everything up to at least 3.12), with venv/pip, and VSCode (with black for formatting and pylance for typechecking), and deployment is github->AWS CodePipeline->AWS Lambda for most things (and everything else is tools for the team where deployment is "clone the repo".)
If I had my choice, I'd switch environment and package management to uv.
VS Code and local Python virtual environments.
Packages managed with pip and versions pinned with pipcompile. One ML production environment forces us to use conda we inherit from the pip environment and try to have as much done by pip as possible.
Formatted with black and flake8
Unit tests with unittest
Azure Dev Ops for running tests, security screening, and CI
The happy path for deployment is build docker images in AzureDevOps unfortunately we can’t do that for good reasons
I’m not a fan of typed Python, it doesn’t work as smoothly as you expect if you’re coming from a statically typed language.
Be careful enforcing pylint quality. It can be a great way to kill your teams velocity.
I’m describing a boring stack that works. All of these tools are 5+ years old and very well supported.
Not a python developer, just casually curious about the discourse. But I wanted to say that I completely agree with the point on letting dynamic languages be dynamic. Otherwise, you reduce the benefit of a dynamic language to the point where I feel like the value proposition comes too close to disappearing. Really at that point, you're just trading performance for compile time or lack there of.
The "basics":
When I need a debugger or a Git GUI, VScode, but Zed is becoming more and more interesting. These support LLM interaction, but things like Claude Code fit better with the above list.Python 3.10+ via VSCode (with GitHub Copilot + Claude Code), Poetry, Pyenv, and Docker
I have been forced into VS Code due to a heavy handed push to use more AI stuff and Copilot in VS Code being the blessed way to do that.
I haven't tried it but I'd assume copilot just exposes an OpenAI compatible API. The Llama.cpp plugin for neovim should work with that OOTB.
I only use it for local models and don't have any hosted ones that I'm willing to pay for so YMMV.
Copilot works with various backend LLMs, including OpenAI, Anthropic, Google, and a few others I think.
We authenticate through a GitHub via SSO, which has a whole RBAC model behind it.
Maybe it would work with something else, but it’s usually more trouble than it’s worth, and depending on the source of the plugin for the other editors, could set off alarms with the security organization, and maybe legal as well. We have a lot of rules around our use of AI.
Can you use vscode devcontainers?
Can you use a dev container with a dockerfile and use UV for package management, ruff for formatting and lint, mypy for type hint checking, and pytest.
And put all of that on your CI as well so if they fail you don’t merge failing code.
I can't answer OP as I only write Python for personal use, not for a "big company".
That said, Neovim + basedpyright + pipenv works well for me.
A lot of people here are using uv and pyenv. Can anyone hint at why I might want to switch?
I'm sorry, can't suggest much, I just use `Neovim` and `ripgrep` search. No AI, no autocompletion, no fancy plugins. At the moment I'm working for one of the biggest Python codebases: Odoo - a general purpose ERP from Belgium.
What do you mean by “tool”? Do you mean IDE/editor, or something else?
By that I meant everything related to python development starting from python version, package manager, environment manager, IDE and ending with deployment tools
This is a great question.
Whatever is sanctioned by the IT department and has been hammered into the permission system. Red tape all around.
And when a new enterprise tool is purchased it may or may not change.
(I only believed it once it was lived through.)
So what tools were sanctioned by IT?
Sorry "working with LLMs" is not enough context to compare .NET and Python environments without resorting to LLMs. People actually use for what?
By that I meant everything related to python development starting from python version, package manager, environment manager, IDE and ending with deployment tools
Where I work Spyder 3.x pipenv for environments Python 3.10 +
Zed is worth a look for your IDe and plays nicely with LLMs
* pyenv for python version management
* poetry for package manager
* vscode + copilot
Black is the main thing you want IMO.
I think language servers and autocomplete are much more useful for .net (I used to do a lot of C# as a freelancer in college) than they are for python. Personally I just open a REPL and tab around in that if I need it but I do that maybe once a month. Most software is well organized such that you can keep enough of the theory in your head for the default nvim config to work easily. A small FITM model with a ring context like Qwen can help a bit with boilerplate if you really want autocomplete.
Python just isn't that verbose.
Ruff has replaced Black imo. Don’t use black in 2025.