Behind the Commit

In this upcoming episode, I chat with Maria Ashna, PyPI support specialist at the Python Software Foundation. We discuss what it's actually like to support nearly a million Python developers with a team of four, the edge cases nobody thinks about, and what PyPI was called before it was PyPI.

Full episode coming out soon!

Antonio Cuni (principal engineer at Anaconda, author of SPy, developer of PyScript and PyPy, co-founder of the HPy project, and creator of PDB++, Fancy Completer, and VMProf) shares why Python is slow and how SPy, a new statically compiled variant of Python, aims to be as fast as C while staying as Pythonic as Python. We discuss the trade-off between dynamic features and performance, how SPy's "red and blue" code model replaces best-effort JIT with predictable errors, why PyPy struggles with C extensions, and what "Pythonic" really means.

Outline

00:00 - Episode highlights and introduction

00:40 - Why is Python so slow?

01:31 - Would a static Python be fast?

02:13 - What is SPy?

02:28 - What motivated you to create SPy? (Spoiler: frustration)

03:52 - Which ideas from PyPy and HPy went into SPy?

06:36 - Biggest challenges of building SPy

07:49 - What does "Pythonic" actually mean?

10:39 - Current roadmap and the two-level language idea

12:24 - Walking through the SPy compilation pipeline

13:57 - Red-shifting and the blue/red AST nodes

14:40 - Why blue and red? (No, it's not from The Matrix)

15:52 - PyPy vs SPy: when to use which

19:34 - C extensions and the C API problem

20:04 - How to actually make your Python code faster

23:34 - Memory management and garbage collection in SPy

24:44 - When will SPy be production-ready?

25:44 - How to contribute to SPy

26:26 - Where does the name "SPy" come from?

Note: the episode was recorded in July 2025 and some features, such as list type, garbage collector, and documentation, have been implemented in the meantime. In the section below you can find the links to those features.

Episode links

– SPy on GitHub: https://github.com/spylang/spy

– PyPy: https://pypy.org/

– HPy project: https://hpyproject.org/

– PyScript: https://pyscript.net/

– Anaconda: https://www.anaconda.com/

– decorators in SPy: https://github.com/spylang/spy/pull/225

– SPy roadmap: https://github.com/spylang/spy/blob/main/ROADMAP.md

– list type in SPy: https://github.com/spylang/spy/blob/main/stdlib/_list.spy

– garbage collector in SPy: https://github.com/spylang/spy/pull/390

– SPy documentation: https://spylang.github.io/docs/dev/

– series of blog posts on SPy: https://antocuni.eu/2025/10/29/inside-spy-part-1-motivations-and-goals/

In this upcoming episode, I chat with Antonio Cuni, the creator of SPy and long-time PyPy developer. We discuss why Python is slow, how a statically compiled Python variant can be as fast as C while still feeling Pythonic, and the "red and blue" model behind SPy's compilation pipeline.

Full episode coming out soon!

In this episode, I'm chatting with Anthony Sottile — creator of pre-commit, primary maintainer of flake8, core contributor to pytest, and maintainer of around 80 open source projects across the Python ecosystem. He's also a GitHub Star and a popular live coding streamer on Twitch under the name "anthonywritescode". We dig into how he actually manages all of it, the origin story of pre-commit, the psychological side of open source maintenance, and how to get started contributing.

Outline

00:00 Episode highlights & Intro

0:59 The all-repos tool — distributed refactoring across repos

2:04 Where the idea came from (Yelp's microservices explosion)

2:42 Tools for managing multiple repositories

3:34 How pre-commit got started (a college group project)

4:15 Rewriting pre-commit for Yelp in 2018

4:46 Hardest technical challenge: supporting 13 programming languages

6:07 Surprising bugs found in NPM and Git

7:05 GitHub Stars and open source funding

8:10 How Sentry approaches funding open source

8:43 The psychological challenges of open source maintenance

10:06 What would you tell your past self?

11:32 How to start contributing to open source

13:05 Why Anthony started streaming on Twitch

13:52 What motivates him to keep streaming

14:58 Has community interaction changed how you design code?

15:48 Where to find Anthony online

Episode links

– pre-commit: https://pre-commit.com

– all-repos: https://github.com/asottile/all-repos

– Anthony's YouTube: https://www.youtube.com/@anthonywritescode

– Anthony's Twitch: https://www.twitch.tv/anthonywritescode

In this upcoming episode, I chat with Anthony Sottile — creator of pre-commit and maintainer of around 80 open source projects. We talk about how he actually manages it all, the surprising bugs he's found in NPM and Git along the way, and the psychological side of open source that nobody talks about.

Armin Ronacher (creator of Flask, previously Sentry’s VP of Platform, and currently founder of a startup Earendil) shares his experience building a startup where 90% of the code is AI-generated. We discuss which programming languages work best with AI agents, why Python's ecosystem makes life harder for AI, and what skills programmers need to stay relevant in the age of AI.

Outline

00:00 Episode highlights and introduction

0:43 - How much code do you write yourself vs AI agents?

2:03 - What kind of problems are suitable for AI and what do you solve yourself?

4:02 - Why do AI agents work better with certain languages like Go vs Python?

7:15 - How to steer AI agents in certain directions?

12:01 - What patterns can AI agents handle well?

15:28 - When starting a new project, what language do you use now?

16:27 - Do you monitor agents and what safeguards do you have?

18:48 - How do you handle parallelization with multiple agents?

19:34 - How should we handle licenses for AI-generated open source libraries?

24:07 - What is the future of programming jobs?

26:31 - What skills should programmers learn to stay competitive?

31:05 - Tips on how to get started with AI agents?

33:44 - How to stay up to date with all the recent changes?

36:16 - Where can people find you online?

Episode links

– Claude Code: https://claude.ai/

– UV (Python package manager): https://github.com/astral-sh/uv

– Simon Willison's blog: https://simonwillison.net/

– Armin's blog: https://lucumr.pocoo.org/

In this upcoming episode, I chat with Armin Ronacher, the creator of Flask. We discuss which programming languages work best with AI agents, why Python's ecosystem makes life harder for both humans and AI, and what skills programmers need to stay relevant in the age of AI.

In this episode, I talk with Mateusz Modrzejewski, a professional musician and AI researcher and assistant professor at the Warsaw University of Technology, about AI-generated music: how it works, what artists think about it, the big copyright questions, and where AI tools can genuinely support the creative process instead of replacing it.

Outline

00:00 Episode highlights and introduction

00:49 Thoughts on AI-generated music

02:09 The good sides of AI

03:40 The bad sides of AI

04:43 How do professional musicians feel?

08:03 The data scraping problem and data poisoning research

11:33 How AI music generation actually works

15:04 How to start experimenting with AI music

18:26 Future of the music industry

25:22 The AI song contest

29:15 Resources to learn more

🎙️ This episode was recorded live at the Venture café at PyWaw in Warsaw.

Episode links

• AI song contest: https://www.aisongcontest.com/
• Librosa library: https://github.com/librosa/librosa
• Pedalboard library: https://github.com/spotify/pedalboard
• Intelligent Instruments Lab: https://iil.is/
• PanGenerator: https://pangenerator.com/
• Pyo library: https://github.com/belangeo/pyo
• ISMIR tutorials: https://ismir.net/resources/tutorials/
• Beyond supervised learning book: https://music-classification.github.io/tutorial/landing-page.html
• Fundamentals of Music Processing book: https://www.researchgate.net/publication/288774112_Fundamentals_of_Music_Processing
• NIME conference: https://www.nime.org/
• ZAiKS Lab https://www.zaikslab.tech
• ⁠Warsaw University of Technology https://www.pw.edu.pl

AI-generated music is on the rise, but what does that mean from a technical and creative perspective? In this upcoming episode, I talk with Mateusz Modrzejewski, a professional musician and AI researcher, about AI-generated music: how it works, what artists think about it, the big copyright questions, and where AI tools can genuinely support the creative process instead of replacing it.

In this episode, I’m chatting with Thibaud Colas — Product Lead & Engineering Manager at Torchbox, Wagtail product lead, and current President of the Django Software Foundation. We talk about digital sustainability: why the internet’s energy use matters, how to measure it, why performance is important for emissions, whether rewriting everything in Rust is the only solution, and why we should keep AI businesses more accountable.

Outline

0:00 — Episode highlights

0:58 — Guest intro

1:21 — Is carbon footprint a popular topic among developers?

3:44 — What’s the carbon impact of the web and software in general?

5:35 — How can we measure a website’s carbon footprint?

8:26 — What are Django’s and Wagtail’s plans to reduce their footprint?

9:35 — Does the programming language we use (e.g., Python or Rust) make a difference?

11:42 — What practical steps can developers take to lower their software’s impact?

12:25 — How do we raise awareness and make sustainability a mainstream topic in tech?

15:34 — Should businesses be accountable for their software’s footprint?

16:59 — What should AI companies and cloud providers be more transparent about?

Episode links

– Sustainable Web Design (methodology): https://sustainablewebdesign.org/

– Wagtail CMS (Torchbox): https://wagtail.org/

– Django Software Foundation: https://www.djangoproject.com/foundation/

– Climate Action Tech community: https://climateaction.tech/

– W3C work on web sustainability: https://w3c.github.io/sustainableweb-wsg/