Python Trending Weekly
ISSUE 110

JIT Compiler Two-Year Retrospective, AI Agent Tools Explosion

2025-07-12 Season 4

TL;DR

A two-year retrospective on CPython's JIT weighs the good, the bad, and the ugly; types are quietly transforming Python, 'PyTorch is dead, long live JAX' ignites debate, and agent tools multiply overnight.

Articles 12 Projects 12
🦄

Articles & Tutorials

× 12
01 Reflections on 2 years of CPython's JIT Compiler: The good, the bad, the ugly fidget-spinner.github.io

Ken Jin, a core developer of CPython's JIT compiler, reflects on two years of development, candidly analyzing the project's successes and challenges. He highlights how the JIT successfully adopted a community-driven development model, attracting numerous contributors. However, performance remains disappointing—often slower than the interpreter under modern compilers. Version 3.15 might finally deliver the real performance gains we've been waiting for.

02 Types are Transforming Python spin.atomicobject.com

This piece traces Python's type system evolution, explaining the major shifts since the typing module landed in Python 3.5. It spotlights NamedTuple and dataclass syntax, showing how class-based type creation became more intuitive and elegant. The article also covers advanced features like generics, NewType, and TypedDict. Combined with tools like pyright and Pylance, the type system isn't just improving code quality—it's fundamentally transforming how we write Python.

03 Solving Wordle with uv's dependency resolver mildbyte.xyz

This clever hack transforms Wordle into a Python package dependency problem, using thousands of virtual package versions to represent letter position constraints and leveraging uv's blazing-fast dependency resolution to automatically solve puzzles. Who knew dependency managers could have such unexpected uses? (Previously, we shared solving Sudoku with Python package resolvers)

04 Developing a Real-time Dashboard with FastAPI, MongoDB, and WebSockets testdriven.io

How do you build a real-time order management dashboard? This tutorial uses MongoDB Change Streams to monitor data changes, combines FastAPI WebSockets for real-time frontend-backend communication, and leverages the Motor async driver for performance gains. It's a solid showcase of modern web app real-time data processing best practices.

05 Escaping Contravariance Hell labs.quansight.org

Don't panic when your type checker screams about "violating the Liskov substitution principle." This article brilliantly explains contravariance using vegetable peelers: a carrot peeler can only peel carrots—you can't use it as a generic vegetable peeler. The solution? Generic protocols and TypeVar to let the type system know exactly which peeler works with which veggie, avoiding the temptation to just slap Any everywhere and call it a day.

06 LangFlow Tutorial: Building Production-Ready AI Applications With Visual Workflows firecrawl.dev

LangFlow makes AI app development as easy as playing with Legos! Drag and drop pre-built components to create complex workflows, building multi-agent systems and RAG applications without writing a single line of code. This tutorial walks through building a Firecrawl search component, covering everything from environment setup to component development to workflow integration.

07 Scaling recommendations service at OLX tech.olx.com

The OLX team shares how they scaled their FastAPI recommendation service to handle tens of thousands of requests per second with p99 latency under 10ms. Through async I/O optimization, batched queries to reduce fan-out, removing Pydantic validation overhead, and tuning garbage collection parameters, they achieved a 40% reduction in latency and 50% boost in throughput.

08 Getting extensions to work with free-threaded Python lwn.net

Python 3.13's free-threaded mode ditches the GIL, but how do you get existing extensions to play nice? The Quansight Labs and Meta teams share their ecosystem migration experience—the key is handling global state races. Using mutexes to protect shared variables, atomic operations to avoid deadlocks, redesigning mutable data structures, and tools like ThreadSanitizer to catch data races, Python is finally ready to embrace the multi-threaded era.

09 Augmented Coding: Beyond the Vibes tidyfirst.substack.com

Through hands-on experience building a B+ Tree library, the author contrasts "Augmented Coding" with "Vibe Coding." The former emphasizes human-AI collaboration where developers stay in the driver's seat, using TDD and clean code principles to guide AI toward generating high-quality code that's more reliable and maintainable.

10 Context Engineering for Agents rlancemartin.github.io

Context engineering has suddenly exploded in popularity. This article breaks it down into four key strategies: write (save outside the context window), select (pull in relevant info), compress (keep essential tokens), and isolate (split processing). Through practical examples like sticky notes, memory systems, and multi-agent architectures, it shows how to elegantly solve context management challenges in long conversations.

Context engineering
11 PyTorch is dead. Long live JAX. neel04.github.io

The author delivers a scathing critique of PyTorch's design philosophy, pointing out how its pursuit of flexibility creates performance bottlenecks at scale. PyTorch tries to have its cake and eat it too—balancing dynamic features with compiler optimization—leading to serious feature fragmentation. JAX, on the other hand, was built around the XLA compiler from day one, using functional programming and pure function constraints to achieve more elegant distributed training and automatic parallelization.

12 operators, not users and programmers jyn.dev

From the "Computers of the Next 200 Years" series, the author drops a bombshell: the distinction between "users" and "programmers" is harmful and kills people's agency. Using spreadsheets as a success story, they show how malleable software can make programming as natural as writing a resume. The article envisions future computing systems with hot reloading, automatic persistence, distributed version control, instant testing, and continuous deployment—turning everyone into computer "operators."

🐿️

Projects & Resources

× 12
01 trae-agent: LLM-based agent for general purpose software engineering tasks github.com

ByteDance's open-source AI coding assistant that supports multiple LLMs and offers a powerful CLI interface with interactive modes. Features a rich tool ecosystem including file editing, bash execution, trajectory recording, and more, built with a modular architecture that's perfect for research and extension. (7.5K stars)

02 xai-sdk-python: The official Python SDK for the xAI API github.com

xAI's official Python SDK built on gRPC, supporting both sync and async clients. Packed with features like multi-turn conversations, streaming responses, image understanding, function calling, and structured outputs. Designed specifically for Python 3.10+ with intuitive APIs and comprehensive documentation.

03 desto: web-interface and cli to manage python and shell scripts in tmux sessions github.com

Manage Python and shell scripts in tmux sessions through a clean web dashboard and CLI. One-click session launching, monitoring, and stopping, plus real-time system status, script chaining, scheduled execution, and live log viewing. Makes script management simple and intuitive.

04 devin.cursorrules: Magic to turn Cursor/Windsurf as 90% of Devin github.com

Transform your $20 Cursor into a Devin-level AI coding assistant through custom config files and Python scripts. Features automated planning and self-evolution, extended tool usage (web browsing, search queries), multi-agent collaboration, and more. Makes AI behave like an intern—planning tasks, updating progress, and learning from feedback. (5.8K stars)

05 LingChat: GPT chat with emotional expressions github.com

An emotional AI chat companion with custom character support and permanent memory. Built-in AI emotion recognition makes expressions change with mood for an immersive chat experience. Includes Vits voice synthesis, visual perception, RAG memory system, and archiving features. Add background music for a truly personalized chat experience.

06 rowboat: AI-powered multi-agent builder github.com

Build multi-agent workflows with AI through a conversational interface for rapid agent assistant creation. Supports MCP server connections, API and Python SDK integration, RAG document uploads, URL scraping, and more. (3.3K stars)

07 agent-factory: A tool for generating Python code for agentic workflows github.com

Mozilla AI's open-source tool for generating agentic workflow code with automated evaluation and testing capabilities. Includes Chainlit multi-turn conversation interface, API services, and web UI—a powerful toolkit for building agent workflows.

08 throttled-py: High-performance Python rate limiting library github.com

High-performance Python rate limiting library supporting multiple algorithms (fixed window, sliding window, token bucket, leaky bucket, and GCRA) and storage backends (Redis, in-memory). Async-ready with support for function calls, decorators, and context managers.

09 tubearchivist: Your self hosted YouTube media server github.com

Self-hosted YouTube video archiving and management system with channel subscriptions, automatic video downloads, indexing, and search capabilities. Web interface for organizing and playing offline videos, integrated with yt-dlp downloader and watch history tracking. (6.5K stars)

10 mlx-audio: A text-to-speech (TTS), speech-to-text (STT) and speech-to-speech (STS) library built on Apple's MLX framework github.com

Text-to-speech, speech-to-text, and speech-to-speech library built on Apple's MLX framework. Supports multiple languages, voice customization, speed adjustment, with an interactive web interface and 3D audio visualization. (2.5K stars)

11 sloppy-xml-py: A Python library to parse malformed XML github.com

Python XML parsing library designed specifically for malformed XML, prioritizing fault tolerance over strict XML compliance. Perfect for parsing LLM-generated XML, offering both streaming and tree-building APIs with robust error recovery and detailed diagnostics.

12 Python-Scripts: A curated list of python scripts for automating your tasks github.com

Curated collection of Python automation scripts covering file management, image processing, data conversion, networking tools, and more. Includes file organizers, image compression, CSV converters, email automation, password generators, and other practical utilities.

ESC
↑↓ select · ↵ open175 issues