Python when possible. Often performance limitations can be handled via good algorithm selection, and careful use of the standard library/other libraries that use C under the hood. If not possible to reach needed performance via Python, look at Mojo, which may become the default choice.

• poetry to install dependencies and handle virtual environments.
• pyenv to manage different versions of python, and avoid using the system python.
• pipx to install and run standalone python cli applications.

• ruff for python linting, auto-formatting.
• Django for full-featured backends.
• FastAPI for standalone/simple APIs.
• Pydantic for data validation.
• HTTPX for making HTTP requests.
• Pytest for testing.
• htmx for reactive web UIs without javascript single-page apps.
• Hypothesis for property-based testing/fuzzing.
• Boto3 for interacting with AWS.
• Polars for data analysis/manipulation.
• Numpy for scientific computing.
• tqdm for displaying progress bars.
• matplotlib for plotting.
• Dash for data visualization via web apps.
• Datasette for data exploration/publishing.
• Scrapy for web crawling/scraping.
• BeautifulSoup for HTML parsing.
• Pillow for image manipulation.
• Pygame for 2d games.
• Rich for rich text formatting in terminals.
• Textual for TUIs (text-based user interfaces).
• Ortools for combinatorial optimization algorithms.
• ScikitLearn for machine learning.
• PyTorch for deep learning.

• PostgreSQL as default option.
• sqlite for simpler projects.
• redis as cache layer if needed.
• JSON as interchange format.

• neovim as primary editor.
• Pycharm for graphical debugging.
• git for source control.
• unix tools combined with pipes (|) to process text generally (awk, sed, wc, grep, tr, diff, uniq, sort, cut, cat, head, tail)
• ripgrep as an improvement to grep.
• fd as an improvement to find.
• jq to parse and process JSON.
• parallel for parallel execution on a single machine.
• iterm terminal emulator.
• zsh shell.
• Excalidraw for diagram creation.

• Chunk functionality together, as our working memories can handle roughly seven items.
• Solve a related, but simpler problem.
• State the problem clearly, and then develop incrementally.
• Repeat tasks manually until patterns emerge and functions discover themselves.
• Build classes independently and let inheritance discover itself.
• Consider data structures programming as a graph-traversal problem, traveling from one "island" to another to use appropriate functionality. (strings -> lists -> dictionaries etc.)
• Separate ETL (extract-transform-load) from analysis. Separate analysis from presentation.
• Verify type and size of the data. View and test a subset of the data.
• Humans should never gaze upon unsorted data.

Adapted from The Mental Game of Python Talk given by Raymond Hettinger at PyBay2019