I've been using Python for over a decade, cycling through virtualenv, pip-tools, poetry, conda (then mamba), and probably some others I'm forgetting. Each promised to solve Python's packaging complexity, but these were all tools I had to fight against (I'm looking at you conda).

Thankfully, we now have uv. We've adopted it at my workplace, transitioning our major repos from poetry. The process was seamless, and our CI build systems are now substantially faster. I've also been adopting it for all of my personal projects.

Migration caveat: When transitioning from poetry to uv, be aware that the dependency solvers use different resolution logic. Even with identical pinned package versions in your pyproject.toml, the resulting lockfiles will contain different transitive dependencies. This can introduce subtle bugs or behavioral changes, so testing is essential during migration.

Mass adoption

The large scale adoption of uv by Python devs is not surprising, it is simply better than the alternatives. Virtual env creation and package installations feel nearly instant. This is a massive boost over what we were dealing with previously.

One of my favourite examples is the wonderfilly wasteful uv run command - we can spin up a disposable Python interpreter, then throw it away in just a few seconds. uv run --with requests script.py creates an isolated environment, installs the package, runs your code, and discards everything—all faster than manually activating and installation dependencies in a traditional venv.

Examples

Starting a new project with uv is easy. If you'd like to follow along, get uv on your system with curl -LsSf https://astral.sh/uv/install.sh | sh.

First, the disposable environment pattern changes how you approach one-off tasks:

# pull up the aws s3 cp commands help menu
uv run --with awscli aws s3 cp help

Creating new projects around uv is easy and the de facto way I'll be setting up my projects in the future.

# Initialize a new project
uv init my-project
cd my-project

# Add (grouped) dependencies directly to pyproject.toml
uv add requests polars
uv add --dev pytest black

# Run your code with all dependencies available
uv run main.py

The generated pyproject.toml is clean and standard:

[project]
name = "my-project"
version = "0.1.0"
requires-python = ">=3.12"
dependencies = [
    "polars>=1.31.0",
    "requests>=2.32.4",
]

[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"

For existing projects, uv provides a drop-in replacement for pip. This is useful for migrating legacy projects:

# Normal pip workflow
pip install -r requirements.txt

# uv equivalent (10-100x faster)
uv pip install -r requirements.txt

Now we just need to wait and see how astral decides to monetize their business — hopefully sanely.