Python Weekly - Issue 769
Python Weekly (Issue 769 October 29 2026)
In partnership with
Welcome to issue 769 of Python Weekly. We have a packed issue this week. Enjoy it!
The playbook for how staff engineers actually operate
You're already doing staff-level work. Someone with half your skill just got the title instead.
The difference isn't the code. It's knowing how the role actually works.
Sign up for The Code newsletter and get:
The complete staff engineer playbook, 6 modules and 25 lessons on scope, sponsors, and the promotion packet that proves you're ready
The Code newsletter (5 min daily) so you keep leveling up. The playbook is just the beginning.
News
CPython's experimental JIT compiler, introduced in Python 3.13, is likely to be removed from the main development branch after its maintainers withdrew PEP 836. The decision follows concerns from Python's Steering Council about performance targets, compatibility, security, and long-term maintenance, leaving the future of the JIT project uncertain.
Microsoft and Meta partnered to introduce the Type Server Protocol (TSP), an open standard decoupling Pylance from Pyright to allow interchangeable type-checking backends in VS Code. This integration lets Python developers switch to Meta's high-speed Pyrefly checker while retaining Pylance's IDE tooling and GitHub Copilot features.
Python's Pillow library fixed a critical-rated vulnerability (CVE-2026-112900) in EPS image handling that could allow remote code execution through malicious filenames passed to Ghostscript. Upgrading to Pillow 13.0.0 addresses this issue alongside several memory corruption and buffer overflow vulnerabilities affecting JPEG and TIFF processing.
Articles, Tutorials and Talks
William Woodruff explains how Astral rebuilt uv's tar parser in Rust to eliminate archive parsing inconsistencies that could expose Python package installations to supply-chain attacks. The new tar-codec uses strict format-specific state machines and safer extraction mechanisms while delivering roughly twice the parsing performance of its predecessor.
This video introduces Jev, a specialized decision model designed to streamline agent workflows by providing structured, fast, and cost-effective outputs for classification tasks rather than generating free-form text. It demonstrates how to integrate Jev as a tool within AI agent architectures to handle conditional logic, filtering, and image analysis more efficiently than traditional LLMs.
Eight videos and over 11 hours of coding, covering evaluation, inference-time scaling, reinforcement learning, and distillation in Python and PyTorch.
Meta Engineering's results on the state of Python typing: AI tools and the checker landscape, what developers actually use and where the friction is.
CF Bolz-Tereick demonstrates a Python-based compiler optimizer that tracks the origin of individual bits through shifts and bitwise operations, enabling simplifications without extensive peephole rewrite rules. Experiments with PyPy and Pydrofoil JIT traces show that bit provenance provides additional optimization opportunities for integer-heavy workloads, although it offers no further reductions in the tested PyPy traces.
Oleg Żero introduces Gnomon, an ontology-based Python compiler that translates natural-language system descriptions into gnomocode to produce validated architecture diagrams. By pairing a deterministic, rule-based validator with an LLM translator, Gnomon detects structural contradictions and generates multiple architectural views from a shared, validated system model.
A security researcher discovered a sandbox escape in AWS's managed MCP Server by exploiting Python's match/case pattern matching and __match_args__ to bypass restrictions on sensitive attribute access. The exploit enabled system command execution and exposed AWS Lambda execution-role credentials, but additional security controls limited its impact, with no access to customer data or other sessions found during testing.
Security researcher Ben Zimmermann scanned 7.25 million public repositories on China's Gitee platform, uncovering 26,400 unique, verified-live credentials, including thousands of cloud and AI service API keys. Using a Python-based serverless pipeline with TruffleHog and Kingfisher, he overcame restrictive search APIs and completed the month-long scan for approximately $3,100.
Drew Breunig explores how longer context windows and persistent AI agents introduce problems such as context poisoning, instruction conflicts, and agents getting stuck pursuing failed approaches. Using Python and DSPy examples, he demonstrates context pruning, tool selection, subagents, and summarization, explaining why periodically resetting an agent's context can improve performance.
The author explains how Apache Arrow provides a shared, in-memory columnar format that bridges file formats like Parquet with high-performance execution engines. By storing data in contiguous, typed memory buffers instead of pointer-heavy language objects, Arrow reduces serialization overhead and improves processing efficiency through CPU cache locality, SIMD instructions, and zero-copy data sharing.
The author details how he trained a 3.8-billion-parameter language model on 65 billion tokens in 43 hours using eight NVIDIA B200 GPUs for approximately $998. Through experiments with FP8 precision, optimizers, model architecture, and training configurations, he demonstrates how to improve GPU efficiency and model performance while documenting approaches that failed to deliver gains.
This video demonstrates how to build Jev, a high-speed inference framework that replaces slow autoregressive decoding with a single prefill-only forward pass for optimized decision-making. It explores implementing this architecture using both encoder and decoder models in PyTorch, covering techniques like causal attention masking, KV-cache reuse, and reinforcement learning for calibrated decision outputs.
Robert Alexander examines why UUIDv7 does not guarantee strictly increasing identifiers, accurate creation timestamps, or unguessable IDs, despite common assumptions about the format. Using Python 3.14's uuid.uuid7(), he demonstrates how a backward system clock adjustment can cause embedded timestamps to drift increasingly far into the future under sustained load.
Implementing soft deletion in Django involves more than adding a deleted_at field, particularly when dealing with bulk deletes, unique constraints, and related objects. This tutorial builds a reusable solution using custom QuerySets, managers, and abstract models, with practical guidance on restoration, cascading behavior, admin integration, and permanent cleanup.
Johnny Metz demonstrates how Django's bulk operations silently bypass model save signals, potentially leaving search indexes, caches, and audit logs out of sync with the database. He introduces a custom QuerySet that blocks these operations by default, using Python's ContextVar and a context manager to permit intentional exceptions while also protecting less obvious ORM write paths.
This video explores various architectural approaches to handling operations on document nodes, comparing traditional methods, the Visitor pattern, and dispatch matrices. It ultimately recommends using Python's singledispatch for a cleaner, more flexible way to manage independent models and operations.
How to use nanodjango, django-nanopages and django-distill to create a static site builder in 3 lines of Python.
The article explains how Peter Hoffmann built an efficient, mobile-friendly map of 33,437 Alpine peaks using Python, JavaScript, and Leaflet, without a database or application server.
Interesting Projects, Tools, and Libraries
Google's Elephant-Goldfish Model (EGM) is an open-source AI-assisted software design workflow that uses context-rich “Elephant” agents to draft specifications and fresh-context “Goldfish” agents to independently review them for missing assumptions, design flaws, and implementation risks.
Debloat any Android or Google TV over Wi-Fi and get an Apple TV-style home.
DeepTutor is an agent-native learning workspace that connects tutoring, problem solving, quiz generation, research, visualization, and mastery practice in one extensible system.
A free AI assistant and survival library that runs on your own computer, with no internet needed. Shows the source of every answer.
A WebAssembly-based Python sandbox for agent-generated code, with the scientific stack included.
Small Python is a lightweight, embeddable Python 3.15 interpreter written in C11 that supports core language features, isolated execution, and compilation of Python scripts into standalone C programs without requiring CPython or external dependencies.
Open-source Python framework to build, orchestrate and deploy AI agents: multi-provider, tool calling, RAG memory and desktop distribution.
Ploom is a Python concurrency testing library that systematically explores thread execution orders to uncover hard-to-reproduce bugs, saving failing schedules as replayable traces and integrating with pytest.
New Releases
These security releases address the compromised eMudhra intermediate CA keys. These keys are now blocked by the ssl module as a last resort against their use. Normal uses of pip, including to upgrade pip itself, are now safe from attacks based on those compromised keys, even if pip hasn’t been updated yet.
Upcoming Events and Webinars
There will be following talks
Squeezing Intelligence out of a Fixed Compute Budget for ARC-AGI-3
Snakes, Arrows and Other Curious Games in Apache Fluss
There will be following talks
How to accept that AI writes better Python code than you do
Harnessing Pydantic AI for Offensive Security
There will be following talks
Data Design Practices for EHR Data
The Shape of the Break: What drives payment gaps in Medicare
There will be a talk, How to create your first AI agent.
Our Other Newsletters
Programmer Weekly - A free weekly newsletter for programmers.
Founder Weekly - A free weekly newsletter for entrepreneurs featuring best curated content, must read articles, how to guides, tips and tricks, resources, events and more.
Update your email preferences or unsubscribe here
© 2026 Python Weekly
228 Park Ave S, #29976, New York, New York 10003, United States
Powered by beehiiv