Talk Python To Me

When you pip install a package with compiled code, the wheel you get is built for CPU features from 2009. Want newer optimizations like AVX2? Your installer has no way to ask for them. GPU support? You're on your own configuring special index URLs. The result is fat binaries, nearly gigabyte-sized wheels, and install pages that read like puzzle books. A coalition from NVIDIA, Astral, and QuantSight has been working on Wheel Next: A set of PEPs that let packages declare what hardware they need and let installers like uv pick the right build automatically. Just uv pip install torch and it works. I sit down with Jonathan Dekhtiar from NVIDIA, Ralf Gommers from QuantSight and the NumPy and SciPy teams, and Charlie Marsh, founder of Astral and creator of uv, to dig into all of it.

Episode sponsors

Sentry Error Monitoring, Code talkpython26

Temporal

Talk Python Courses

Links from the show

Guests

Charlie Marsh: github.com

Ralf Gommers: github.com

Jonathan Dekhtiar: github.com

CPU dispatcher: numpy.org

build options: numpy.org

Red Hat RHEL: www.redhat.com

Red Hat RHEL AI: www.redhat.com

RedHats presentation: wheelnext.dev

CUDA release: developer.nvidia.com

requires a PEP: discuss.python.org

WheelNext: wheelnext.dev

Github repo: github.com

PEP 817: peps.python.org

PEP 825: discuss.python.org

uv: docs.astral.sh

A variant-enabled build of uv: astral.sh

pyx: astral.sh

pypackaging-native: pypackaging-native.github.io

PEP 784: peps.python.org

Watch this episode on YouTube: youtube.com

Episode #544 deep-dive: talkpython.fm/544

Episode transcripts: talkpython.fm

Theme Song: Developer Rap

🥁 Served in a Flask 🎸: talkpython.fm/flasksong

---== Don't be a stranger ==---

YouTube: youtube.com/@talkpython

Bluesky: @talkpython.fm

Mastodon: @[email protected]

X.com: @talkpython

Michael on Bluesky: @mkennedy.codes

Michael on Mastodon: @[email protected]

Michael on X.com: @mkennedy

When you type a question into ChatGPT, the model only has what you typed to work with. But tools like Claude Code can plan, iterate, test, and recover from mistakes. They work more like we do. The difference is the agent harness: Planning tools, file system access, sub-agents, and carefully crafted system prompts that turn a raw LLM into something genuinely capable.



Sydney Runkle is back on Talk Python representing LangChain and their new open source library, Deep Agents: A framework for building your own deep agents with plain Python functions, middleware hooks, and MCP support. This is how the magic works under the hood.

Episode sponsors

Sentry Error Monitoring, Code talkpython26

Agentic AI Course

Talk Python Courses

Links from the show

Guest

Sydney Runkle: github.com

Claude Code uses: x.com

Deep Research: openai.com

Manus: manus.im

Blog post announcement: blog.langchain.com

Claudes system prompt: github.com

sub agents: docs.anthropic.com

the quick start: docs.langchain.com

CLIs: github.com

Talk Python's CLI: talkpython.fm

custom tools: docs.langchain.com

DeepAgents Examples: github.com

Custom Middleware: docs.langchain.com

Built in middleware: docs.langchain.com

Improving Deep Agents with harness engineering: blog.langchain.com

Prebuilt middleware: docs.langchain.com

Watch this episode on YouTube: youtube.com

Episode #543 deep-dive: talkpython.fm/543

Episode transcripts: talkpython.fm

Theme Song: Developer Rap

🥁 Served in a Flask 🎸: talkpython.fm/flasksong

---== Don't be a stranger ==---

YouTube: youtube.com/@talkpython

Bluesky: @talkpython.fm

Mastodon: @[email protected]

X.com: @talkpython

Michael on Bluesky: @mkennedy.codes

Michael on Mastodon: @[email protected]

Michael on X.com: @mkennedy

If you've built documentation in the Python ecosystem, chances are you've used Martin Donath's work. His Material for MKDocs powers docs for FastAPI, uv, AWS, OpenAI, and tens of thousands of other projects. But when MKDocs 2.0 took a direction that would break Material and 300 ecosystem plugins, Martin went back to the drawing board. The result is Zensical: A new static site generator with a Rust core, differential builds in milliseconds instead of minutes, and a migration path designed to bring the whole community along.

Episode sponsors

Sentry Error Monitoring, Code talkpython26

Talk Python Courses

Links from the show

Guest

Martin Donath: github.com

Zensical: zensical.org

Material for MkDocs: squidfunk.github.io

Getting Started: zensical.org

Github pages: docs.github.com

Cloudflare pages: pages.cloudflare.com

Michaels Example: gist.github.com

Material for MkDocs: zensical.org

gohugo.io/content-management/shortcodes: gohugo.io

a sense of size of the project: blobs.talkpython.fm

Zensical Spark: zensical.org

Watch this episode on YouTube: youtube.com

Episode #542 deep-dive: talkpython.fm/542

Episode transcripts: talkpython.fm

Theme Song: Developer Rap

🥁 Served in a Flask 🎸: talkpython.fm/flasksong

---== Don't be a stranger ==---

YouTube: youtube.com/@talkpython

Bluesky: @talkpython.fm

Mastodon: @[email protected]

X.com: @talkpython

Michael on Bluesky: @mkennedy.codes

Michael on Mastodon: @[email protected]

Michael on X.com: @mkennedy

When LLMs write code to accomplish a task, that code has to actually run somewhere. And right now, the options aren't great. Spin up a sandboxed container and you're paying a full second of cold start overhead plus the complexity of another service. Let the LLM loose on your actual machine and... well, you'd better be watching.



On this episode, I sit down with Samuel Colvin, creator of Pydantic, now at 10 billion downloads, to explore Monty, a Python interpreter written from scratch in Rust, purpose-built to run LLM-generated code. It starts in microseconds, is completely sandboxed by design, and can even serialize its entire state to a database and resume later. We dig into why this deliberately limited interpreter might be exactly what the AI agent era needs.

Episode sponsors

Talk Python Courses

Python in Production

Links from the show

Guest

Samuel Colvin: github.com

CPython: github.com

IronPython: ironpython.net

Jython: www.jython.org

Pyodide: pyodide.com

monty: github.com

Pydantic AI: pydantic.dev

Python AI conference: pyai.events

bashkit: github.com

just-bash: github.com

Narwhals: narwhals-dev.github.io

Polars: pola.rs

Strands Agents: aws.amazon.com

Subscribe Running Pydantic’s Monty Rust sandboxed Python subset in WebAssembly: simonwillison.net

Rust Python: github.com

Valgrind: valgrind.org

Cod Speed: codspeed.io

Watch this episode on YouTube: youtube.com

Episode #541 deep-dive: talkpython.fm/541

Episode transcripts: talkpython.fm

Theme Song: Developer Rap

🥁 Served in a Flask 🎸: talkpython.fm/flasksong

---== Don't be a stranger ==---

YouTube: youtube.com/@talkpython

Bluesky: @talkpython.fm

Mastodon: @[email protected]

X.com: @talkpython

Michael on Bluesky: @mkennedy.codes

Michael on Mastodon: @[email protected]

Michael on X.com: @mkennedy

Monorepos -- you've heard the talks, you've read the blog posts, maybe you've seen a few tantalizing glimpses into how Google or Meta organize their massive codebases. But it's often in the abstract and behind closed doors. What if you could crack open a real, production monorepo, one with over a million lines of Python and over 100 of sub-packages, and actually see how it's built, step by step, using modern tools and standards? That's exactly what Apache Airflow gives us.



On this episode, I sit down with Jarek Potiuk and Amogh Desai, two of Airflow's top contributors, to go inside one of the largest open-source Python monorepos in the world and learn how they manage it with uv, pyproject.toml, and the latest packaging standards, so you can apply those same patterns to your own projects.

Episode sponsors

Agentic AI Course

Python in Production

Talk Python Courses

Links from the show

Guests

Amogh Desai: github.com

Jarek's GitHub: github.com

definition of a monorepo: monorepo.tools

airflow: airflow.apache.org

Activity: github.com

OpenAI: airflowsummit.org

Part 1. Pains of big modular Python projects: medium.com

Part 2. Modern Python packaging standards and tools for monorepos: medium.com

Part 3. Monorepo on steroids - modular prek hooks: medium.com

Part 4. Shared “static” libraries in Airflow monorepo: medium.com

PEP-440: peps.python.org

PEP-517: peps.python.org

PEP-518: peps.python.org

PEP-566: peps.python.org

PEP-561: peps.python.org

PEP-660: peps.python.org

PEP-621: peps.python.org

PEP-685: peps.python.org

PEP-723: peps.python.org

PEP-735: peps.python.org

uv: docs.astral.sh

uv workspaces: blobs.talkpython.fm

prek.j178.dev: prek.j178.dev

your presentation at FOSDEM26: fosdem.org

Tallyman: github.com

Watch this episode on YouTube: youtube.com

Episode #540 deep-dive: talkpython.fm/540

Episode transcripts: talkpython.fm

Theme Song: Developer Rap

🥁 Served in a Flask 🎸: talkpython.fm/flasksong

---== Don't be a stranger ==---

YouTube: youtube.com/@talkpython

Bluesky: @talkpython.fm

Mastodon: @[email protected]

X.com: @talkpython

Michael on Bluesky: @mkennedy.codes

Michael on Mastodon: @[email protected]

Michael on X.com: @mkennedy