TheNote
Planet Python Note
GooglePlay AppStore

Planet Python

The website Planet Python is a planet site that aggregates Python-related content from various sources, including blogs, news sites, and other online publications. The website provides a one-stop destination for individuals to stay up-to-date with the latest developments in the Python programming world. The content on the site includes tutorials, news, project announcements, and discussions about various Python-related topics. Users can visit the site to stay informed about the Python community, new releases, conferences, and best practices in using the Python programming language. The website's purpose is to help promote and disseminate Python-related content, thereby contributing to the growth and development of the Python community.
Planet Python planetpython.org
RSS planetpython.org
RSS Hunter RSS Hunter • Aug 19, 2024

Thread Of Notes

Armin Ronacher: Anger, Anxiety and Agency

Anger at work is generally unproductive and creates a negative environment for colleagues. Discovering a shared vision within a company and aligning with it, or accepting it if disagreement exists and change isn't possible, is crucial. A prevalent question in the tech industry is how to avoid anger amidst rapid changes, particularly concerning AI. The author suggests that feelings of disorientation and anxiety are more appropriate responses than anger. Anxiety acknowledges uncertainty about the future without assigning blame. Anger, however, requires a target and implies external wrongdoing. While AI may bring productivity gains, there's concern these will benefit companies over employees. Many tech leaders express doubts about AI, worried about costs and competition from large AI labs. Instead of anger, the author proposes embracing uncertainty, which can foster curiosity and engagement with new developments. Genuine excitement is another alternative, leading to a sense of empowerment and freedom. The author notes that AI's benefits are often seen in personal projects, not just company profits. This shift highlights the uncertainty leaders face, with many projecting confidence despite private doubts. The author experiences this duality of excitement and uncertainty regarding their own profession and business. While anger can feel more actionable by providing a villain, it often misdirects focus during disruptive changes like AI. Although societal, environmental, and global impacts of these changes are concerning, dwelling on anger towards individuals is unproductive. Instead, remaining curious and excited about experimentation is encouraged. This approach will allow for informed decisions about when and where resistance is truly warranted.
https://lucumr.pocoo.org/2026/8/24/anger-anxiety-agency/ lucumr.pocoo.org
RSS Hunter RSS Hunter • Today

LernerPython blog, from Reuven Lerner: AI seminars — just one of the big changes at LernerPython

The author has been busy improving LernerPython based on global feedback, leading to several significant updates. A monthly AI seminar is being added to LernerPython memberships to address the growing importance of AI for developers. The first seminar, scheduled for August 24th, will focus on using Claude Code as a supershell. Additionally, an AI-powered Socratic tutor trained on the author's extensive materials is now live on LernerPython.com. This tutor is available to help users with questions in their native language. Older issues of Bamboo Weekly will now be freely accessible after two years, providing hundreds of data analysis exercises. The author has also created new documentation for Pandas methods used in Bamboo Weekly, featuring practical examples and common mistakes. These new Pandas method explanations are available even to non-subscribers. A comprehensive calendar of upcoming office hours and a library of past recorded sessions are now accessible on the LernerPython.com events page. These changes aim to enhance learning and provide more accessible resources for Python and AI development.
https://lernerpython.com/2026/08/23/ai-seminars-just-one-of-the-big-changes-at-lernerpython/ lernerpython.com
RSS Hunter RSS Hunter • Yesterday

Ned Batchelder: Micro language implementation: Calcium

I wrote a tiny language implementation: Calcium. It’s meant as a demonstration of how languages like Python are implemented. It has a tokenizer, a parser, an AST, a compiler, bytecodes, and an execution engine, all in about 300 lines of code. I did it because I often see the question: isn’t Python interpreted? Why do people say it’s compiled? (BTW, I also answered this in an earlier blog post: Is Python interpreted or compiled? Yes.) It can be hard to explain that your Python program never becomes an explicit sequence of native CPU instructions, which is what people often think “compiled” means. So I coded up Calcium to have on hand the next time it comes up. I think it will help to be able to show the execution engine code reading bytecodes and doing what they say. It could also be an interesting starting point for people wanting to play with a language implementation. It has almost nothing, so there’s lots of simple things (comments?) to add.
https://nedbatchelder.com/blog/202608/micro_language_implementation_calcium nedbatchelder.com
RSS Hunter RSS Hunter • Aug 22

Robin Wilson: Some recent, and not so recent, talks and activities

The author is catching up on blog posts after a hiatus. They recently had a talk accepted for FOSS4G UK 2026, which prompted them to write about a workshop they led at FOSS4G UK 2025. The workshop focused on generating live vector tiles from a PostGIS database using the pg_tileserv tool. Participants learned how to serve tiles directly from tables and PostGIS functions, with practical exercises provided. The workshop aimed to demonstrate the power of on-the-fly vector tile generation for complex geospatial workflows.Separately, the author also presented a talk titled "Lights, Spectrometer, Action!" at the Electromagnetic Field festival. This talk explored the science of spectrometry, involving live demonstrations with a spectrometer. They examined light from various sources and its interaction with different materials. The talk also touched upon satellite imaging applications of spectrometry and potential uses in chemistry.Looking ahead, the author will present at FOSS4G UK 2026 on making Environment Agency DEM data accessible. They have converted this data into cloud-native formats like STAC and COG, facilitating easier access and analysis. This work allows for efficient creation of DEM mosaics and custom web map visualizations. The author is also developing custom software for spectrometer control.
https://blog.rtwilson.com/some-recent-and-not-so-recent-talks-and-activities/ blog.rtwilson.com
CdXz5zHNQW_3HN87m7Gll.png
RSS Hunter RSS Hunter • Aug 22

Armin Ronacher: Fast and Hard Code

The rise of LLMs has significantly streamlined the process of programming, making language choice less critical for developers. The ability to rewrite code in different languages or generate code in unfamiliar ones reduces friction for humans. This shift allows language selection to be driven more by marketing and perceived benefits, such as performance. For example, Rust has seen increased adoption, partly due to a growing desire for faster software and LLMs' ability to optimize code efficiently. Experts like Mitchell Hashimoto and others, known for their focus on performance, are also embracing agent-written code. Tools like autoresearch enable developers to achieve optimization without needing extensive prior knowledge.Beyond Rust, other "hard languages" like Zig are also gaining traction in projects aiming for speed and small footprints. Cloudflare's Artifacts service and Vercel's fx agent, both utilizing Zig, exemplify this trend, often with LLM assistance. Furthermore, developers are now tackling traditionally challenging technologies like DWARF files, eBPF, custom network drivers, and even old computing hardware. Many of these areas were previously inaccessible or intentionally gatekept. While this change might introduce some "slop," it also brings more developers into the fold who value speed and efficiency, potentially leading to a new era of highly performant software.
https://lucumr.pocoo.org/2026/8/22/fast-hard-code/ lucumr.pocoo.org
RSS Hunter RSS Hunter • Aug 22

Bob Belderbos: Unsubscribe links without a login: Django signing

Django's signing module allows secure unsubscribe links without user login or sessions. An insecure approach directly using user IDs in URLs is vulnerable to mass unsubscribing. Storing random tokens in the database is a common but more complex fix. Django's django.core.signing module provides tamper-evident tokens without requiring database storage by signing a value (like a user's primary key) with a SECRET_KEY.A "salt" is crucial to prevent token collisions between different features, ensuring a token for one purpose (e.g., announcements) cannot be used for another (e.g., forum notifications). While signed, tokens are not secret; the data within them is readable but cannot be altered without invalidating the signature. Therefore, do not sign sensitive information meant to be hidden. Crucially, unsubscribe actions should happen on a POST request, not a GET request, to prevent mail scanners and prefetchers from unintentionally unsubscribing users. A confirmation page on GET, followed by a POST to finalize the action, is the recommended flow.Unsubscribe tokens often do not require expiry, as clicking an old link twice yields the same safe outcome. However, for sensitive actions like magic login links, max_age argument in signing.loads provides expiry without database storage. For single-use token scenarios, like email verification, additional state management (database record of used tokens) is necessary since the signature alone cannot track usage.
https://belderbos.dev/blog/unsubscribe-without-login-django-signing/ belderbos.dev
CdXz5zHNQW_kaNMCAHtWT.webp
RSS Hunter RSS Hunter • Aug 22

Caktus Consulting Group: Fuzzy String Matching in Django and PostgreSQL

Gerald Carlton and I will be presenting on fuzzy name search at DjangoCon US 2026 on Monday, August 24, and this is the companion blog post. Searching for a person by name is harder than it looks: names might be typed differently by different operators or change over a person’s lifetime; for example, Smith could be entered as Smyth, Smythe, or Smidt, and Weiss as Weiß. Although names are particularly susceptible to misspelling, these strategies apply to all fuzzy string matching.
https://www.caktusgroup.com/blog/2026/08/21/fuzzy-string-matching-django-postgresql/ caktusgroup.com
RSS Hunter RSS Hunter • Aug 21

Django Weblog: DSF Membership Open Space at DjangoCon US

DjangoCon US attendees are invited to a 45-minute open space session about the Django Software Foundation. This event will take place on Wednesday, August 26th, from 1:00 pm to 1:45 pm in the Wolf Point Ballroom. DSF Board Directors and other DSF representatives will be present for open discussion and to answer questions. Key topics will include fundraising efforts and the $500,000 goal and its funding, particularly for Django Fellows. The ongoing search for the DSF's first Executive Director will also be discussed, with applications closing on September 14, 2026. Information on individual membership, how to become a member, and the benefits of membership, such as voting rights, will be shared. The session will also cover working groups and teams, clarifying that membership is not required to join. Updates on DEP 19, concerning technical governance, and DEP 20, regarding Django's release schedule, will be provided. This open space is intended for all attendees, regardless of their current DSF membership status, to ask questions about the Foundation's operations and decision-making. Questions about the Foundation itself, not specific code issues, are encouraged. For those unable to attend, questions can be submitted via the "Contact the DSF" page or through the Django Forum.
https://www.djangoproject.com/weblog/2026/aug/20/dsf-membership-open-space-at-djangocon-us/ djangoproject.com
RSS Hunter RSS Hunter • Aug 21

The No Title® Tech Blog: Optimize Images X adds internationalization

Optimize Images X, the multi-platform desktop application that helps you reduce the file size of your images on macOS, Windows and Linux, has just been updated to version 2.2.0. This new version adds full internationalization support with some initial translations, plus a handful of minor UI adjustments.
https://no-title.victordomingos.com/articles/2026/optimize-images-x-internationalization no-title.victordomingos.com
RSS Hunter RSS Hunter • Aug 20

PyCharm: PyCharm for AI-assisted Django Workflows

A recent survey indicates AI is integral to most developers' workflows, with 90% using it weekly or daily. While AI can write code rapidly, developers remain responsible for understanding, evaluating, and approving its output. This underscores the crucial role of the IDE, with PyCharm highlighted as a tool that enhances developer control. PyCharm allows users to select their preferred AI agents and models, offering native support for leading options and an extensive registry for more. Users can opt for a JetBrains AI subscription or leverage their own tools, including local models via Ollama or LM Studio. The IDE also empowers developers to train AI agents on specific project conventions and coding standards through reusable "Skills." PyCharm provides robust, version-specific support for Django, including its latest features, ensuring alignment even when AI training data is outdated. Furthermore, the IDE offers comprehensive tools for code review, such as visual diffs and an integrated Git history, alongside a local history feature for tracking changes independently of version control. PyCharm's Django Logical Structure view visualizes the application's architecture from the framework's perspective, aiding comprehension. Finally, its Endpoints tool window and data editor/viewer facilitate API verification and database interaction, allowing developers to confidently assess AI-generated code and its impact.
https://blog.jetbrains.com/pycharm/2026/08/pycharm-for-ai-assisted-django-workflows/ blog.jetbrains.com
RSS Hunter RSS Hunter • Aug 20

Talk Python to Me: #559: 12 Things You Should (and Shouldn't) Do in AWS

Matt Lea, an expert in managing costly hourly outages, emphasizes that crucial decisions for handling emergencies are made long before they occur. He advocates for proactive measures taken during ordinary workdays to prevent future problems. The discussion covers his top twelve AWS best practices, such as using infrastructure as code and IAM roles. He also addresses the applicability of these practices for individuals managing smaller VPS instances. The conversation then shifts to Cloud War Games, a platform where deliberate disruptions are simulated. This allows teams to gain practical incident response experience without real-world consequences. These simulations are designed to be their first real incident within a controlled environment. The ultimate goal is to equip teams with the preparedness needed for unexpected system failures. This preparation helps minimize the impact and reputational damage of an outage.
https://talkpython.fm/episodes/show/559/12-things-you-should-and-shouldnt-do-in-aws talkpython.fm
RSS Hunter RSS Hunter • Aug 19

Anarcat: The people vs the AI overlords

Large Language Models (LLMs) are exacerbating an existing crisis within the open-source software community, particularly regarding the volume of code reviews. This pressure extends beyond open-source, impacting the entire computing field. LLMs are fundamentally changing how programming is done, even for those who try to avoid them. Users are employing LLMs to automate extensive code reviews, generating significant compute costs.This automation raises concerns about the potential for API pricing changes, essentially charging for what was previously more accessible. The author, a proponent of free and open-source software, observes a shift towards renting software, mirroring a past era where acquiring tools like compilers was a significant expense. Now, many software tools and LLM access require ongoing monthly payments, leading to a situation where developers must pay to get their work done.This trend is seen as "enshitification," where companies train models on stolen work and then sell it back to developers. Furthermore, LLMs are presented as a threat to engineering jobs, contributing to a deteriorating job market with fewer opportunities and lower pay. Tech companies are perceived as actively attacking the engineering profession by pushing for the use of models they control.Even without mandatory LLM use, the pressure to keep pace with LLM-generated work is immense, potentially compromising supply chain security. The author contends that the LLM hype is part of a broader societal attack on workers and the environment. Addressing this requires collective political and social action, rather than individual adaptation.
https://anarc.at/blog/2026-08-18-people-vs-ai-overlords/ anarc.at
RSS Hunter RSS Hunter • Aug 19

PyCharm: What’s Fixed and Improved in PyCharm 2026.2

PyCharm 2026.2 introduces 263 fixes and improvements focused on enhancing Python code insight. Key updates include better type inference, reduced false positives, and more reliable refactoring. The release offers improved support for SQLAlchemy 2.0, resolving numerous previously problematic scenarios. Code insight benefits from fixes in control-flow narrowing and handling of unreachable code. Strings within type annotations and iterable unpacking now see more accurate analysis. Augmented assignments and constructor return types are also analyzed more precisely. Enum members now yield Literal types for their value and name attributes. Parameter types are now inferred from decorators, and completion and auto-import have been made smarter. PyCharm now prioritizes reusing existing imports and supports auto-importing nested classes. Completion for unittest.mock.patch targets is now functional, and overriding built-in methods fills in complete annotated signatures. Type inlay hints display inferred type arguments inline, and f-string format-spec validation has been extended. The Rename refactoring now correctly updates module references, and the Field refactoring is now termed Attribute. Overall, these enhancements lead to a more precise and predictable understanding of Python code within PyCharm.
https://blog.jetbrains.com/pycharm/2026/08/what-s-fixed-and-improved-in-pycharm-2026-2/ blog.jetbrains.com
RSS Hunter RSS Hunter • Aug 19

Python Software Foundation: How AWS Powers PyPI and the PSF

The Python Software Foundation (PSF) manages critical infrastructure like PyPI, Python.org, and CPython, supporting Python's widespread use across various sectors. Most of the PSF's infrastructure runs on AWS, with significant costs covered by AWS's Open Source Credits Program. PyPI handles billions of requests daily, with Fastly's edge caching reducing the need to hit AWS origin servers. However, uncached interactions like user uploads are managed by services like EC2, RDS, and S3. This year, AWS spending has notably increased by 69%, a departure from eight years of stable costs. This rise is attributed to increased usage from more users, automated agents installing packages, and numerous CI runs. The PSF emphasizes efficient credit utilization, advocating for upstream fixes in popular tools like GitHub Actions. A major ongoing project is migrating the entire fleet to Amazon EKS for better autoscaling. Currently, infrastructure work is heavily reliant on a small team, highlighting a critical need for more engineering staff. The PSF is seeking funding through PyPI Organization accounts and service agreements to enhance its services and mitigate supply chain risks.
https://pyfound.blogspot.com/2026/08/how-aws-powers-pypi-and-psf.html pyfound.blogspot.com
RSS Hunter RSS Hunter • Aug 19

Python GUIs: Creating PySide6 UI without .ui / Qt Designer — Build your entire GUI in pure Python code, no .ui files required

It is entirely possible to create PySide6 applications without relying on .ui files or Qt Designer. Developers can construct all GUI elements, including windows, buttons, and labels, directly within Python code. This approach offers greater control and keeps all interface logic consolidated. A basic PySide6 application requires a QApplication instance and a QWidget to display. Widgets can be created and parented to a window, but without layouts, they occupy fixed positions. Qt layouts like QVBoxLayout, QHBoxLayout, and QGridLayout are essential for automatically arranging widgets. To manage complexity, it's recommended to subclass QWidget or QMainWindow. Signals and slots enable interactivity; for instance, a button's 'clicked' signal can be connected to a Python function. QMainWindow provides a framework for menus, toolbars, and status bars, with content placed in a central widget. Nested layouts can be achieved by adding one layout to another using addLayout(). The demonstration culminates in a complete QMainWindow application, showcasing programmatic GUI creation with interactive elements and a status bar.
https://www.pythonguis.com/faq/creating-pyside2-ui-without-ui-qt-designer/ pythonguis.com
RSS Hunter RSS Hunter • Aug 19

Armin Ronacher: What Is Reasoning

A recent paper revealed how to extract reasoning traces from closed-weight models, sparking online discussion. Reasoning traces are typically hidden from users, unlike in open-weight models where they are visible. These traces are essentially text that models are trained to output onto a scratchpad before their final answer. Special tokens demarcate these reasoning sections, with a parser directing this content to a separate stream. For closed models, this intermediate text is likely redacted or summarized. The amount of reasoning a model performs is determined by its system prompt, not sampling. For instance, it might be set to "low" or "Absolute maximum with no shortcuts permitted." Models are trained to keep this scratch work separate from the final response channel. Tricking a model into believing it is within the reasoning channel can cause these traces to be leaked. Older models, when thinking was disabled, sometimes wrote their thoughts to a null device. In some cases, the only special behavior is the model's ability to refrain from thinking. This can be achieved by disabling the model's usual thinking mechanisms. Some inference APIs might prefill tokens to manage reasoning, which can be bypassed by custom tools, particularly when native reasoning is disabled. The author humorously encountered content filters while trying to use a specific model for grammar checking this very blog post.
https://lucumr.pocoo.org/2026/8/19/what-is-reasoning/ lucumr.pocoo.org
CdXz5zHNQW_KvYYV2hcVr.png
RSS Hunter RSS Hunter • Aug 19

PyCoder’s Weekly: Issue #748: Line Breaks, OpenCode, Segfaults, and More (2026-08-18)

This issue of PyCoder's Weekly covers various Python topics, starting with the complexity of splitting multi-line text due to Unicode. A Real Python course introduces AI-assisted Python coding using OpenCode and Gemini. PR-AF is highlighted as a top open-source code reviewer, outperforming commercial tools. Two lines of Python code are noted to cause a segfault in the interpreter due to lazy annotations. Election candidates for the Packaging Council and the PSF Board for 2026 were announced. Django is moving to an annual release cycle, and new Python 3.12.14, 3.11.16, and 3.10.21 versions have been released, along with Django Rest Framework 3.18.0. Articles discuss OAuth 2.0 authorization code flow, the cognitive overhead of pandas, and integrating OpenTelemetry with FastAPI. New projects include pytest-leak-finder, num2words, marimo-book, pyrig, and cog. Several Python conferences and meetups are scheduled throughout August.
https://pycoders.com/issues/748 pycoders.com
CdXz5zHNQW_XCM4q7iYG0.png
RSS Hunter RSS Hunter • Aug 18

Python Bytes: #492 Codeberg Puts Head in Sand

Python 3.12.14, 3.11.16, and 3.10.21 have been released as security-fix-only updates. These releases address multiple critical vulnerabilities including tarfile hardening, path-traversal bypasses, and various CVEs related to Python's standard library. Users on these versions, especially those handling untrusted tarballs, are strongly advised to update.Codeberg's recent ban on projects primarily generated by AI is being examined for its implications as a potential GitHub alternative. This policy raises questions about whether code hosting platforms should regulate the creation process of software versus focusing on its behavior and impact. The difficulty in defining and enforcing "mostly generated" code presents a significant challenge to this approach.Brett Cannon has highlighted missing components for achieving reproducible builds on PyPI. Currently, there's no standardized method to record the exact source code a distribution originated from or the specific build tools used. Implementing solutions for these gaps would involve modifying sdist metadata and potentially introducing an sdist version 2.A major focus in reproducible builds is ensuring the integrity of the software supply chain. The goal is to make reproduction a seamless process for creators while empowering build backends and installers with the necessary information. This could lead to PyPI displaying independently reproduced builds, offering enhanced trust for users.Pydantic Logfire is sponsoring the episode, offering observability for AI applications. It provides unified tracing across agents, LLMs, APIs, and databases, down to the infrastructure level. Logfire utilizes OpenTelemetry and allows querying all data with Postgres-compatible SQL.The sponsor emphasizes that even AI applications are fundamentally engineering efforts. They offer a free tier and an easy onboarding process for developers to integrate Logfire into their applications.In other news, uv has transitioned to preferring post-quantum key exchange for enhanced security in its operations. This update reflects a growing trend towards incorporating future-proof cryptographic methods.Other "extra" items include upgrades to MCP servers, syncing agentsview via postgres, new offerings for Talk Python courses, and the release of the "Lean TDD" audiobook. The episode concludes with a brief joke about a dog.
https://pythonbytes.fm/episodes/show/492/codeberg-puts-head-in-sand pythonbytes.fm
RSS Hunter RSS Hunter • Aug 18

Seth Michael Larson: When str.lower() is a security vulnerability in Python

Internet standards often lack support for non-ASCII characters, necessitating a mapping for domain names. NamePrep, defined in RFC 3491, was designed for this purpose as part of IDNA 2003. This older IDNA standard has since been superseded by IDNA 2008, specified in RFCs 5890-5893. Python offers support for both versions, with IDNA 2003 accessible via the idna codec and IDNA 2008 via the external idna package. The stringprep module in Python's standard library implements StringPrep. A crucial step in StringPrep is "case folding," which involves mapping characters to lowercase or uppercase for case-insensitive comparisons. This process utilizes mapping tables B.2 and B.3 from RFC 3454, which are based on Unicode 3.2.0 rules. A vulnerability was identified in Python's implementation where the standard str.lower() function, using newer Unicode versions, deviated from the required Unicode 3.2.0 case-folding rules. This inconsistency resulted in different IDNA encodings for certain characters. The fix involved creating specific exceptions to ensure Python's str.lower() behavior aligns with Unicode 3.2.0 for StringPrep and IDNA calculations. This remediation ensures IDNA 2003's consistency with its specification.
https://sethmlarson.dev/when-str-lower-is-a-security-vulnerability sethmlarson.dev
RSS Hunter RSS Hunter • Aug 18

Bob Belderbos: Guardrails Protect Your Codebase. What Protects Your Judgment?

Cal Newport warns that AI-driven speed in coding may erode the critical thinking skills that define good engineering. Instead of actively understanding code, developers risk passive recognition, becoming detached from the foundational understanding. While some advocate for a complete return to manual coding, the author seeks a balanced approach. This middle ground involves intentionally integrating friction, which is crucial for learning and skill development.Skill atrophy occurs not from using AI itself, but from delegating the internal modeling process. The author distinguishes between skill atrophy and AI's potential for generating subtly incorrect code. To maintain ownership and understanding, developers should re-derive complex code paths and ensure they can explain every change. Narrowing AI tasks to well-defined problems helps keep engineers involved and prevents overwhelming code reviews.Corey Schafer's methods, like detailed prompts and thorough code review, offer practical ways to preserve this friction. However, human attention is finite, necessitating automation of the remaining tasks. This is where "correctness by design" becomes vital, employing stricter system-level checks like strong typing and compile-time assertions.Rust exemplifies this by making invalid states unrepresentable, forcing AI compliance. In Python, type checkers and validation models serve a similar purpose. Ultimately, responsible AI use in development requires both retaining control over critical decisions to sharpen judgment and implementing strict rules to ensure AI's accountability. Software engineering remains a process of judgment and experience, with AI automating tedious aspects. The key takeaway is to identify and refuse to delegate the most important decisions to AI, ensuring true ownership and robust code.
https://belderbos.dev/blog/ai-skill-atrophy-keep-the-friction/ belderbos.dev
RSS Hunter RSS Hunter • Aug 18

The Python Coding Stack: Back From Holidays • SOLID Ideas, Agents, Rust

Returning from holiday, the author faces a daunting backlog of personal responsibilities and professional challenges. As a seasoned Python educator, the author acknowledges the profound impact of AI on programming education and the economic landscape of their profession. This realization prompts a commitment to explore this evolving world publicly over the coming months. The author intends to document their process of learning and adapting to AI-driven changes in coding. This exploration will include discussions on how we learn, teach, write, and assess code in an AI-shaped future. A planned series will delve into SOLID programming principles, emphasizing their importance for robust AI-generated code. Another section will unpack AI agents, examining their nature and efficient usage. The author also plans to document their personal journey of learning the Rust programming language. These upcoming articles promise a mix of Python content and adjacent topics. The author invites readers to join them on this learning adventure.
https://www.thepythoncodingstack.com/p/back-from-holidays-solid-ideas-agents thepythoncodingstack.com
RSS Hunter RSS Hunter • Aug 17

Brian Okken: Lean TDD Book Launch

Hey all. I’m super excited to announce the launch of my new book.It’s Lean TDD: TDD Without the Waste.That link in the last sentence is a link to leantdd.com.There you’ll find links to all the places you can read or listen to it.Right now, it’s available at:Amazon as a paperback Amazon as an ebook Audible as an audio book Other platforms to come later …In the next few months, it will get onto other platforms. I’ve already had requests for Kobo, and just raw files sold through my site, and I’d also like to get it onto library platforms.
https://pythontest.com/lean-tdd-launch/ pythontest.com
RSS Hunter RSS Hunter • Aug 17

Brett Cannon: What's missing to have reproducible builds on PyPI

The author highlights the lack of a defined method for reproducible builds in Python packaging. Reproducible builds are crucial for supply chain security, allowing third parties to verify that distribution contents match the source code. This process can detect tampering during the build and identify the use of compromised tools. Even pure Python wheels are vulnerable if their build backends are compromised.To achieve reproducible builds, two key pieces of information are needed: the source code location and the software used in the build process. Currently, source code location is not explicitly recorded in sdists or wheels, though installers sometimes record it. A mechanism to include source location information in distribution metadata is proposed.For recording build software, wheels can utilize Software Bill of Materials (SBOMs), but sdists lack a similar structured format, necessitating a potential new sdist version. The author suggests build backends could record their execution environment, and pip maintainers could assist in logging this information without burdening distribution producers.To make reproducibility visible and beneficial, the idea of trusted verifiers independently reproducing distributions and reporting success to PyPI is proposed. This would allow users to identify verified distributions and benefit from a more secure supply chain. This feature should be presented as an optional perk, not a mandatory requirement, to avoid discouraging users.
https://snarky.ca/whats-missing-to-have-reproducible-builds-on-pypi/ snarky.ca
RSS Hunter RSS Hunter • Aug 16

Juri Pakaste: TIL: Typed do in Swift

The author discovered a valuable feature in Swift's typed throws that simplifies error handling. Initially, they found typed throws very useful for specific error types but encountered a problem where complex code blocks would cause Swift to widen the error type to 'any Error'. This meant their 'catch' block would no longer precisely type the error, leading to compilation issues. Their workaround involved using 'catch let error as MyError', but this necessitated an unreachable catch-all block, making the error-handling incomplete and frustrating. The critical discovery was the 'do throws(MyError)' syntax. By explicitly declaring the expected error type in the 'do' block, Swift's compiler maintains the precise error typing. This elegant solution prevents the compiler from getting confused, ensuring that the 'catch' block correctly handles the specified error type. This feature improves the clarity and completeness of error handling when working with typed throws in Swift. The author's oversight of this detail in the original enhancement proposal was a significant miss. Now, all is well-typed unicorns and sunshine.
https://juripakaste.fi/swift-do-typed/ juripakaste.fi
RSS Hunter RSS Hunter • Aug 15

Python Software Foundation: Announcing the Packaging Council Election Candidates for 2026!

The inaugural Python Packaging Council (PPC) election is underway with seventeen nominees vying for five open seats. The election will fill all five positions on the council. The top two vote-getters will serve two-year terms, designated as Cohort A. The next three highest vote-getters will serve one-year terms, designated as Cohort B. Future elections will alternate which cohort is up for re-election, ensuring staggered turnover. Nominations closed on August 11th, and candidates were announced on August 13th. Voters must affirm their intention to vote by August 25th, 2:00 pm UTC. All PSF Voting Members must affirm their membership to participate. Voting will commence on September 1st and conclude on September 15th. Eligible voters who have affirmed their participation will receive an email with their ballot from OpaVote.
https://pyfound.blogspot.com/2026/08/announcing-packaging-council-election.html pyfound.blogspot.com
RSS Hunter RSS Hunter • Aug 13

Python Software Foundation: Announcing the PSF Board Candidates for 2026!

Eighteen candidates are running for four open seats on the Python Software Foundation Board. Cheuk Ting Ho, Christopher Neugebauer, Denny Perez, and Georgi Ker are completing their current terms. Voters are encouraged to read candidate nomination and supporting statements to understand their priorities and vision. These statements are crucial for making informed decisions as board members influence budget, programs, and PSF direction. The nomination period closed on August 11th, with candidates announced on August 13th. To vote, PSF Voting Members must affirm their intention by August 25th. Voting will take place from September 1st to September 15th. Past voters may be automatically added to the voter roll if their email hasn't changed. Those who affirmed their intention will receive a ballot via email from OpaVote on September 1st. Any questions about membership or the election can be directed to [email protected].
https://pyfound.blogspot.com/2026/08/announcing-psf-board-candidates-for-2026.html pyfound.blogspot.com
RSS Hunter RSS Hunter • Aug 13

Wingware: Wing Python IDE Version 12.0.2 - August 13, 2026

Wing Python IDE version 12.0.2 has been released with several new features and improvements. The release includes a native ARM64 Windows version of Wing, improved remote development on Windows, and enhanced performance and responsiveness. The new version also streamlines Claude Code setup and reduces the size of the analysis cache database by about 20%. Wing 12 integrates the Claude Code AI coding agent directly into the IDE, with a new Claude Code tool and a Tasks tool for planning and reviewing AI agent work. The IDE is available in two feature-based product tiers: Wing Pro, the full-featured Python IDE, and Wing Classic, the complete traditional Python IDE. Wing 12 also includes many other improvements, such as automatic test file discovery, improved source code analysis, and faster detection of externally modified files. The product line has been simplified, with the Commercial and Non-Commercial use distinction replaced by the two feature-based product tiers. Existing users of Wing Personal may continue to use Personal 11.x indefinitely, switch to free Wing 101, or purchase a Wing Classic license. Overall, Wing 12 offers a range of new features and improvements to enhance the development experience, including AI-driven coding tools and improved performance and responsiveness.
https://wingware.com/news/2026-08-13 wingware.com
CdXz5zHNQW_mXxF48doZc.png
RSS Hunter RSS Hunter • Aug 13

Trey Hunner: Reorganizing Python's sys module

The sys module in Python is described as a "junk drawer" holding many miscellaneous items. This article explores what the sys module might look like if designed from scratch today. Currently, sys has over a hundred attributes, many of which are functions, leading to a crowded namespace. To address this, the author proposes reorganizing sys into nine submodules.These proposed submodules include sys.cli for command-line arguments, sys.imports for module-related functions, sys.io for standard input/output, sys.repl for interactive prompt settings, sys.interpreter for system and installation details, sys.memory for memory management tools, sys.exceptions for exception handling, sys.profile for profiling and introspection, and sys.runtime for interpreter behavior. Some attributes within these proposed submodules bear resemblance to existing standard library modules, hinting at potential further reorganization.The article acknowledges the significant challenges of actually implementing such a reorganization. Primarily, the transition period and maintaining backward compatibility for existing code that relies on the current flat namespace of sys present major hurdles. A technical solution involving module-level attribute handling is discussed, demonstrating how the old flat namespace could be preserved for compatibility. However, the author doubts this reorganization will ever happen due to the vast amount of existing code and the Python community's general reluctance to make such foundational changes unless absolutely necessary. The main takeaway for developers is that submodules can be a useful organizational tool for large "utils" modules.
https://treyhunner.com/2026/08/reorganizing-pythons-sys-module/ treyhunner.com
RSS Hunter RSS Hunter • Aug 13

Django Weblog: DSF Office Hours

The Django Software Foundation Board hosts weekly open office hours every Wednesday at 6:00 PM UTC, accessible to all Django community members without the need for an invitation or agenda. These sessions, ongoing since October 2024, offer a direct line of communication with the board and other key figures. Currently, the board is particularly interested in discussing the search for a Django Executive Director. They encourage potential applicants to attend office hours to ask questions about the role and the application process. The foundation has also increased its 2026 fundraising goal to $500,000 to support hiring an Executive Director, requiring approximately $16,000 in monthly recurring support. They are seeking corporate sponsorships and donations to meet this goal. Beyond these specific topics, office hours cover the activities of working groups, ongoing foundation projects, and general DSF operations. Community members can also use this time to provide feedback before board meetings. However, office hours are not intended for general Django support or product marketing. Other communication channels are available for those who cannot attend on Wednesdays.
https://www.djangoproject.com/weblog/2026/aug/12/dsf-office-hours/ djangoproject.com
RSS Hunter RSS Hunter • Aug 12

PyCharm: What’s New in PyCharm 2026.2.1

PyCharm 2026.2 introduces significant AI integration, allowing AI agents to work directly within Jupyter notebooks using a live kernel. This persistence of variables and models improves reliability for machine learning workflows. The release also addresses package management issues by ensuring AI agents install packages into the correct project environments. Marimo notebooks are now supported via a new third-party plugin, enabling reactive cells and interactive UI elements within the IDE. To enhance performance and focus, PyCharm is unbundling less-used plugins like Data Wrangler and Hugging Face. The Python Packages tool window has been redesigned for better dependency visualization and faster package management. Type checking messages are now clearer and more actionable with detailed breakdowns of mismatches. Virtual environment creation for end-of-life Python versions is no longer supported. SQLAlchemy 2.0's Session.get() inference has been improved for better model instance recognition. These updates are available in PyCharm 2026.2.1 and aim to provide a more streamlined and powerful development experience, particularly for AI-focused projects.
https://blog.jetbrains.com/pycharm/2026/08/what-s-new-in-pycharm-2026-2-1/ blog.jetbrains.com
CdXz5zHNQW_bGZrtpxsuT.png
RSS Hunter RSS Hunter • Aug 12

PyCharm: We Stopped AI Agents From Installing Into the Wrong Python: Task Success Rates Jumped to 95%+

AI agents often struggle with Python projects, installing dependencies incorrectly or ignoring virtual environments. This leads to broken setups and wasted developer time. PyCharm's new Agent Environment Coordinator skill dramatically improves AI agent performance in Python development. It allows agents to access and utilize the project's specific Python environment.In tests, this skill boosted average task success rates from 68% to 98% across six AI models and 28 Python tasks. Crucially, these agents no longer pollutes the system Python environment. Previously, AI models failed to recognize project-specific interpreters, leading to global installations and script failures. The Agent Environment Coordinator enables agents to query PyCharm for the correct Python interpreter and its managing tool.If no environment exists, it can configure one using PyCharm's existing mechanisms. The skill provides context without taking control, allowing the agent to construct commands itself. This means agents work with existing project setups out of the box, eliminating the need for manual coaching or cleanup. This functionality is available with a JetBrains AI subscription. The methodology involved 28 everyday Python environment tasks, with success and system cleanliness being key metrics. The results clearly show that AI models simply lacked the necessary context, not capability.
https://blog.jetbrains.com/pycharm/2026/08/we-stopped-ai-agents-from-installing-into-the-wrong-python-task-success-rates-jumped-to-95/ blog.jetbrains.com
RSS Hunter RSS Hunter • Aug 12

PyCharm: We Gave AI Agents a Live Jupyter Kernel in PyCharm

Traditional AI agents struggle with Jupyter notebooks, often corrupting files and losing model state upon completion. A new Jupyter skill integrated into PyCharm addresses this by allowing AI agents to operate within a live Jupyter kernel. This innovation ensures state persistence across cells, preventing notebook corruption. It also optimizes long jobs by having the agent wait for execution rather than constantly polling. This live kernel approach proved to be approximately 12% cheaper for Claude Opus 5 across twelve machine learning tasks. The cost savings stem from an improved cache-warming mechanism, leading to a higher percentage of cache reads. Previously, AI tools treated notebooks as plain text, leading to corruption and loss of critical runtime information when using subprocesses. The new Jupyter skill leverages PyCharm's internal notebook intelligence to provide the agent with direct kernel control. This allows agents to write and run Python code directly, maintaining variables and trained models. The skill also implements a more efficient waiting mechanism and reads only new output, reducing token waste. While cost savings are evident, especially for stateful jobs, the skill also offers workflow improvements for other models. Users are reminded to explicitly instruct agents to save their work. Complex ML tasks may still require human intervention, as the skill addresses tooling inefficiencies, not fundamental ML difficulty. This functionality is available with a JetBrains AI subscription.
https://blog.jetbrains.com/pycharm/2026/08/we-gave-ai-agents-a-live-jupyter-kernel-in-pycharm/ blog.jetbrains.com
RSS Hunter RSS Hunter • Aug 12

PyCharm: Unbundling and Deprecating Low-Usage Plugins in PyCharm

PyCharm will begin unbundling and deprecating low-usage plugins starting with version 2026.2. This change is part of ongoing maintenance to keep PyCharm fast and responsive. Plugins like Data Wrangler, Hugging Face, and Google Colab are among those being affected. The goal is to focus development effort on features with broader impact for Python developers. Unbundling does not mean removing the plugins entirely. Their code will be moved to an Obsolete Plugins repository. Users can still install compatible versions from the JetBrains Marketplace for PyCharm 2026.2. However, the PyCharm team will cease active maintenance and feature development for these plugins. Starting with version 2026.3, compatible versions will no longer be published by the PyCharm team. The community can continue to build and install these plugins manually from the repository. The PyCharm team thanks users for their past contributions to these plugins.
https://blog.jetbrains.com/pycharm/2026/08/unbundling-and-deprecating-low-usage-plugins-in-pycharm/ blog.jetbrains.com
RSS Hunter RSS Hunter • Aug 12

Python GUIs: Adding QTabWidget to a Layout Alongside Other Widgets in PyQt6 — How to combine QTabWidget with other layouts without it taking over your entire window

Placing a QTabWidget alongside other layouts in PyQt is indeed possible without special workarounds. The common issue of the QTabWidget appearing to dominate the layout stems from its tab pages lacking content or their own internal layouts. When tab pages are empty, the QTabWidget might request disproportionate space, causing other widgets to collapse. To resolve this, ensure each tab page includes a layout and some content. Additionally, setting a reasonable size policy or stretch factor for the QTabWidget helps it share space effectively with neighboring widgets. A minimal example demonstrates combining QTabWidget with other stacked widgets in a QHBoxLayout. Stretch factors can be applied to layouts and individual widgets within the main layout to control the distribution of available space. For instance, assigning a higher stretch factor to the QTabWidget gives it more relative space. A complete example illustrates integrating multiple sections, including various layout types and the QTabWidget, all coexisting harmoniously. This approach ensures all components are visible and share the window's dimensions appropriately. Ultimately, the key is to provide structure and content within the QTabWidget's pages.
https://www.pythonguis.com/faq/cant-integrate-qtabwidget-into-a-layout/ pythonguis.com
RSS Hunter RSS Hunter • Aug 12

Python Bytes: #491 Feeling Judged

The episode discusses several significant developments in the Python ecosystem. Claude Code's new /insights feature offers a detailed analysis of user coding habits by processing local session transcripts. Post-quantum cryptography has arrived in Python via the pyca/cryptography package, introducing ML-KEM and ML-DSA standards to a widely used library. The MCP protocol has undergone a major rewrite, moving to a stateless core and renaming FastMCP to MCPServer for improved scalability and efficiency. Inshellisense provides advanced IDE-style command-line autocompletion for over 600 tools across Windows, Linux, and macOS. Xweather is highlighted as a sponsor, offering weather intelligence APIs for AI agents. Additional news includes the release of Django 6.1, an upcoming DjangoCon US, and Michael Kennedy's new Rust course.
https://pythonbytes.fm/episodes/show/491/feeling-judged pythonbytes.fm
RSS Hunter RSS Hunter • Aug 12

PyCoder’s Weekly: Issue #747: Modern OOP, bisect, Django Async, and More (2026-08-11)

Real Python has released a new book, "Modern Object-Oriented Python," focusing on classes and object-oriented design. The bisect module is highlighted for implementing binary search in Python, with examples of its functions. PropelAuth is sponsoring a solution for securely integrating AI agents into B2B applications. Discussions around Django and asynchronous programming reveal significant updates and changes. Several Python Enhancement Proposals (PEPs) are detailed, including drafts for frozen syntax and extensible JSON serialization, and accepted PEPs for yield from in async generators and freezing the HTML Simple Repository API. Python versions 3.14.7 and 3.13.15, as well as Django 6.1, have been released. The Python Typing Survey for 2026 is open for participation, aiming to guide the future of Python's type system. Articles cover using Hydra for modular configuration, the potential for asyncio.all_tasks() to silently drop tasks in free-threaded builds, and advanced Celery recipes. Other topics include programmatic LLM prompt development with DSPy, Django's setUpTestData for faster tests, and thoughts on SIMD in pure Python. A useful tool is introduced that identifies the Python version a feature was added in. Further articles explain data validation with Pointblank and sending emails with Python. Projects like xy, autowt, vscode-marimo, and commerce are featured. A list of upcoming Python and Django events from August 2026 is provided.
https://pycoders.com/issues/747 pycoders.com
CdXz5zHNQW_14ZOYttUhz.png
RSS Hunter RSS Hunter • Aug 11

Brett Cannon: My nomination statement for the 2026 Python packaging council

The author is running for the inaugural Python Packaging Council (PPC) with Microsoft's support, though independently motivated. They acknowledge the challenge of introducing themselves to a broader PSF membership compared to core developers. The author highlights their extensive experience in Python packaging, including over 23 years as a core developer and numerous contributions to packaging PEPs. They have also served on the Python Steering Council and co-maintained the 'packaging' project.As a candidate, they aim to leverage their steering council experience to help establish the new PPC effectively. A key goal is improving the developer experience for both package producers and consumers. This includes clarifying specifications and potentially adopting successful practices from tools like uv that deviate slightly from current specs for efficiency.Furthermore, the author is committed to enhancing the security of the Python supply chain. They propose making reproducible builds easier to allow independent verification of package integrity. The author also sees value in expanding the use of Software Bills of Materials (SBOMs) throughout the packaging process to identify and mitigate vulnerabilities. Their overall vision is to improve packaging security without negatively impacting developer experience.
https://snarky.ca/my-nomination-statement-for-the-2026-python-packaging-council/ snarky.ca
RSS Hunter RSS Hunter • Aug 10

James Bennett: Breaking up (lines) is hard to do

The seemingly simple task of splitting text into lines is surprisingly complex due to historical and evolving standards. Early computing relied on ASCII, which included control characters like LINE FEED (LF) and CARRIAGE RETURN (CR) for line breaks. Different operating systems adopted various combinations of these, leading to portability issues for plain text. For example, Windows uses CR LF, Unix uses LF, and classic Mac OS uses CR. Python addressed this by introducing "universal newline" mode, recognizing all three common line break formats when opening files. Beyond the traditional CR and LF, ASCII also defined FORM FEED (FF) and VERTICAL TAB (VT), which also cause line breaks. The advent of Unicode introduced seven new code points and the CR LF sequence as line break indicators. These include LINE FEED, LINE TABULATION, FORM FEED, CARRIAGE RETURN, NEXT LINE, LINE SEPARATOR, and PARAGRAPH SEPARATOR. Additionally, Unicode's bidirectional algorithm considers characters with a bidirectional class of 'B' as paragraph separators, which also function as line breaks. These include the INFORMATION SEPARATOR FOUR, THREE, and TWO, also known by their ASCII names FILE SEPARATOR, GROUP SEPARATOR, and RECORD SEPARATOR. In total, Python's splitlines() method recognizes ten distinct code points and one multi-code-point sequence as line breaks. This comprehensive set accommodates historical conventions and modern Unicode standards for handling text line divisions.
https://www.b-list.org/weblog/2026/aug/10/newlines/ b-list.org
RSS Hunter RSS Hunter • Aug 10

Talk Python to Me: #558: Hyper-Personal Software with Python

The concept of dark-matter enterprise software refers to internal tools built by companies that are not well understood and are often left untouched. These tools are typically built by individuals who have since left the company, and there is a general reluctance to modify or update them. As a result, they can become frozen in time, with many similar tools existing in the shadows. Michael Booth has written about the idea of hyper-team software, where small teams within large companies build their own tools to fill gaps in functionality. This approach can be beneficial, but it also carries risks if not managed properly. The idea of hyper-team software is an extension of the concept of hyper-personal software, which refers to customized tools built for individual use. Michael Booth's article explores the potential benefits and pitfalls of hyper-team software, including the need for guardrails to prevent chaos. The discussion around hyper-team software is relevant in the context of companies losing significant value due to outdated or poorly maintained internal tools. The topic is also connected to the idea of AI-assisted tools and the potential for small teams to drive innovation within large companies. The conversation around hyper-team software highlights the importance of finding a balance between allowing teams to build their own tools and maintaining a level of oversight and control to prevent problems from arising.
https://talkpython.fm/episodes/show/558/hyper-personal-software-with-python talkpython.fm
RSS Hunter RSS Hunter • Aug 10

Django Weblog: Django is moving to an annual release cycle

Django is transitioning to an annual release cycle starting January 2028, with all feature releases receiving three years of support. This new model aligns better with Python's annual release schedule, ending Django's previous issue of supporting outdated Python versions. Each new Django release will support the three most recent Python versions available at that time. The distinction between regular and Long-Term Support (LTS) releases is being retired, as all feature releases will now have an LTS-level commitment. This change eliminates the need for users to undertake large, infrequent upgrades by allowing them to update annually. Three Django versions will be supported concurrently, providing a predictable target for third-party package developers. API stability and deprecation policies remain consistent, with deprecation periods effectively extended. The first release under this new cycle will be Django 2028. Current support commitments for Django 5.2 LTS and 6.2 LTS will remain in effect. The full details and reasoning are available in DEP 20.
https://www.djangoproject.com/weblog/2026/aug/10/annual-release-cycle/ djangoproject.com
RSS Hunter RSS Hunter • Aug 10

Ned Batchelder: Caller-specific coverage

The author desires to measure code coverage for a function on a per-caller basis. This is motivated by refactoring error-checking code in their BASIC interpreter, Acidica. Initially, each built-in function had its own argument validation, allowing specific coverage of error conditions. After refactoring, this detailed coverage was lost as validation moved to a shared helper function.The proposed solution involves a decorator that leverages coverage.py's dynamic contexts. This decorator creates a new coverage context named after the calling site (file and line number) before executing the decorated function. Upon function completion, the original context is restored.A proof-of-concept decorator using coverage.Coverage.current() and cov.switch_context() demonstrates this functionality. This requires a minor, unreleased change to coverage.py to support nested contexts. The author found that their HTML coverage reports now show distinct contexts for each caller of the expects function, revealing which branches were executed by whom.Future improvements include post-processing these contexts to identify missing coverage for specific callers and enabling hierarchical or multiple contexts to track both caller and execution details simultaneously. The author also aims to integrate this functionality via coverage configuration rather than modifying source code with decorators. Potential applications include using function arguments, like func_name, as context identifiers.
https://nedbatchelder.com/blog/202608/callerspecific_coverage nedbatchelder.com
CdXz5zHNQW_IFrCzZRuFh.png
RSS Hunter RSS Hunter • Aug 9

LernerPython blog, from Reuven Lerner: Free real-world Pandas exercises, with solutions

The author emphasizes that to improve at Pandas, one needs to practice with real-world data, not just tidy tables. The author has written Bamboo Weekly, a weekly set of Pandas exercises based on real, current, public data, for the last 3.5 years. As of now, issues older than two years are available for free, with over 500 exercises and fully worked-out solutions. The author highlights that real data teaches you how to handle messy data, such as badly named columns and inconsistent date formats. The exercises in Bamboo Weekly are designed to mimic real-world problems, requiring multiple steps to solve. The author also provides method guides for 16 common Pandas methods, covering what the method does, examples, and common mistakes. These guides are verified against Pandas 3 and include links to matching exercises on the LernerPython practice system. The author suggests starting with a method guide or an exercise that interests you and trying to solve it before reading the solution. The goal of Bamboo Weekly is to improve your data-analysis muscle memory, making you more confident in tackling problems at work. By practicing with real data and learning from common mistakes, you can become more proficient in using Pandas and improve your overall data analysis skills.
https://lernerpython.com/2026/08/09/free-real-world-pandas-exercises/ lernerpython.com
RSS Hunter RSS Hunter • Aug 9

Ed Crewe: From Routing Checks to Trajectory Testing: Evaluating an Agentic Chatbot

A company is developing an AI test framework for Postgres AI Hybrid Manager, a product that integrates traditional Postgres management with AI tools like Langflow. This chatbot aims to unify product features, help documentation, and AI workflows into a single conversational interface. Testing such an LLM-backed agent is challenging because its responses are non-deterministic and can be subtly wrong even when fluent. The framework evolved to address this by focusing on testing the entire chat trajectory, not just final answers.Key concepts are introduced, including "Goldens" (perfect desired outputs) and "Rubrics" (qualitative checklists for good answers). For AI testing, LLMs are used as judges to score outputs against Goldens or Rubrics, translating complex responses into pass/fail results. Task Completion Rate (TCR) is used to aggregate multiple eval passes. The testing strategy began with simple routing evals, checking if the chatbot directed prompts to the correct specialized agent or skill.As the system evolved from agents per tool to a consolidated orchestrating agent with skills, routing evals adapted to check for the visibility and selection of the right skill. Subsequently, TCR evals were implemented to assess if the chatbot's full response successfully completed the user's task, using LLM-as-a-judge with specific rubrics. This led to two execution modes: Direct mode for prompt and rubric development, and Proxy mode to test the full production path.The framework also had to handle multi-step conversations, where the test unit became the entire conversation, not just single turns. This required preserving conversation state across multiple API calls in Proxy mode or simulating it in Direct mode. Scoring now includes per-step checks and an overall conversation score. The underlying testing library used is deepeval, upon which the company built its pipeline, plugin system, CI integration, and Langfuse push. Routing is implemented as a custom BaseMetric in deepeval, demonstrating how specific business rules can be integrated into the LLM evaluation process.
http://edcrewe.blogspot.com/2026/08/from-routing-checks-to-trajectory.html edcrewe.blogspot.com
CdXz5zHNQW_2wtMsPJlJt.png
RSS Hunter RSS Hunter • Aug 8

Django Weblog: Call for applicants for a Django Executive Director

The Django Software Foundation is seeking an Executive Director, a new paid operational leadership role. This position reports to the Board and is responsible for strategic direction and day-to-day management of the nonprofit. The Executive Director will act as the primary liaison between the Board, staff, community, and sponsors. The DSF, a 501(c)(3) nonprofit, supports the development and promotion of the Django web framework. This role is crucial for maturing the Foundation's structure and ensuring its sustainability. Key responsibilities include fundraising, aiming to increase annual goals, and managing administrative operations. The Executive Director will also oversee the DSF Assistant and Django Fellows program. Public representation, marketing, and outreach to grow awareness of the DSF are also vital duties. The role requires experience in nonprofit leadership, fundraising, operations, and stakeholder communication, with technical knowledge being a plus. The position is remote within the United States and offers a salary range of $90,000 to $120,000, plus a potential bonus. Applications are due by September 14, 2026.
https://www.djangoproject.com/weblog/2026/aug/06/call-for-applicants-for-a-django-executive-directo/ djangoproject.com
RSS Hunter RSS Hunter • Aug 6

Django Weblog: Django 6.1 released

The Django team has released Django 6.1, introducing new features like model field fetch modes, database-level delete options for ForeignKey, and dictionary-based email settings. This release marks the end of mainstream support for Django 6.0, which will receive security and data loss fixes until April 2027. Users are encouraged to upgrade to Django 6.1 to benefit from the latest improvements and continued support. The new version is available for download from the official Django website or the Python Package Index.
https://www.djangoproject.com/weblog/2026/aug/05/django-61-released/ djangoproject.com
RSS Hunter RSS Hunter • Aug 5