Unit 20 · Language valence 2: Python on Hacker News · q27b

It’s wild to think that 5 years ago we were still arguing about whether `dataclasses` were a good idea, and now they’re the default for almost every new project. The ecosystem has matured so much that the "batteries included" philosophy actually feels true again, rather than just a marketing slogan. The real story isn't Python itself, it's the tooling. Ruff replacing flake8/black/isort in a single binary was the turning point. I switched my entire team over last year and our CI times dropped by 40%. If you're still running a separate linter and formatter, you're doing it wrong. I miss the days when "Python in 202X" posts were about web frameworks. Now it’s all AI/ML infrastructure and data engineering. It feels like the language has been hijacked by the GPU boom. Don't get me wrong, PyTorch is great, but I worry we're losing the general-purpose scripting soul that made Python accessible to non-CS majors in the first place. (greedy generation)

I’ve been writing Python since 2016, and honestly, the biggest change isn’t the language features—it’s the expectation of performance. Ten years ago, if your script took 10 seconds, you shrugged. Today, with the rise of LLMs and real-time data pipelines, that 10-second delay is a dealbreaker. We’re seeing a shift where people are no longer just "using Python" but actively managing its bottlenecks. The adoption of `pydantic` for data validation, `polars` over `pandas` for heavy lifting, and the gradual move toward `mypy` strictness in production code shows a maturation. We’re not just scripting anymore; we’re engineering. It’s less about "Python is slow" and more about "know when to drop into Rust/C via bindings." The ecosystem has finally grown up to handle scale without sacrificing readability. (greedy generation)