Great collection of pytest+playwright tips for E2E testing (with a mention of my axe-core wrapper for a11y tests!) www.better-simple.com/django/2025/...

✈️ Headed to Chicago! My agenda: Weds: VS Code Dev Day www.eventbrite.com/e/vs-code-de... Thurs: AI Tour aitour.microsoft.com/flow/microso... Fri: OpenHack ms-workshops.cloudevents.ai/ms-innovatio...

Yesterday was school picture day, and we didn't even have time to get my kid's hair untangled, while other kids had slicked back hair & fancy hairdos. I was so ashamed I was sobbing after getting home from drop-off. I just saw the pic and good news: 1st graders are adorbs even with unbrushed hair! 😅

If anyone needs empowerment to hang up your halloween decorations: I hereby grant you permission. (Just don't hang up the polyester spiderwebs that kill wildlife) https://www.snopes.com/news/2023/10/20/decorative-spider-webs-dangerous-or-no/

My front door camera keeps spotting someone at the front door, while there's noone there... BUT I did just hang up our halloween skeleton decorations for the year! 💀 Oops.

Takeaways from latest TalkPythonToMe podcast: 1) Learn Kubernetes 2) Play with DuckDB 3) Wrap head around free-threading vs async https://www.youtube.com/watch?v=4BQrovxemXY

I woke up last night at 3AM with a jolt, watching an earthquake move through the house. Apparently it was only a 4.3, and the most damage it did was break a few windows, but a surprisingly wide radius of people felt it. www.berkeleyside.org/2025/09/22/b...

If I start collecting the insects that have died (naturally) in my garden, is that weird? Creepy weird or adorable weird? I could make a lil museum for the next garden tour!

Me, trying desperately to exit the Python REPL on Windows. (I'm a Mac user)

Canceled our Disney+ account, after seeing them give in to Trump's unreasonable demands. Our family's a huge fan of Disney movies, but luckily we've got Miyazaki movies on tap.

Disney gets burned

I went hardcore and bought 100% dark chocolate chips and put them in a generic mason jar in the pantry. Should I warn my partner?? Teeheehee.

Why is this such a banger?? https://www.youtube.com/watch?v=KMU0tzLwhbE

I made a new video for the VS Code channel, showing how to use a custom chat mode with MCP servers (GitHub, Playwright) to fix issues in a full-stack web app: www.youtube.com/watch?v=QLAI...

New blog post: Why you should filter the tools from MCP servers, and how to actually do it in VS Code GitHub Copilot, Langchain v1, and Pydantic AI https://blog.pamelafox.org/2025/09/filter-tools-from-mcp-servers.html

I have many repos where I'm the sole maintainer, and it can feel a bit lonely to work on them. I like assigning PRs to @Copilot to review in those repos - I feel less lonely, and I often get some actionable feedback too.

For me, one of the best way to truly understand a framework is to try to write the same thing across multiple frameworks. That forces me to go beyond what's in the tutorial for a framework and discover the constraints of each framework.

Achievement unlocked: rode a nudibranch! Thanks to @simonwillison.net for discovering the Cannonball Arts Center in Seattle - def check it out. It also has a VR golf cart ride on a track which simulates a horror car wash, and the most random gift shop ever.

I'm in Redmond for a few days, and I just used my Microsoft badge for the *first time ever*, after working "here" for 3 years. I have no intention to RTO, but I did enjoy using the whiteboards to sketch out a plan with a colleague. I always loved a good whiteboarding session.

Nice presentation on research that digs deep into how LLMs perform with large amounts of context: hamel.dev/notes/llm/ra... The research goes beyond the "needle in a haystack" benchmark, which is not as useful anymore as it's not measuring the complexity of real RAG scenarios.

LLMs introduce a huge range of new capabilities for research, but also make it possible for researchers to "hack" their results in new ways by how they chose to use models for annotation This is a useful pass at quantifying some of the risk, and some mitigation strategies arxiv.org/pdf/2509.08825

I feel like not enough people realize that skippers are SO FREAKING CUTE. the furbys of the lepidoptera world, i always say.

I've swapped gpt-5 for gpt-oss:120b in my issue researching agent, and well..I really miss gpt-5. The gpt-oss:120b is actually doing a decent job calling research tools, but it seems to forget what it's doing when it's time to summarize the research. 😢

@cameron.pfiffer.org any chance you happen to know what SLM on Ollama both works well with their structured outputs implementation AND is generally high quality? (I want to use gpt-oss but the ollama team has to patch a harmony bug to get struct outputs working)

GitHub issues can now specify what issues they’re blocked by: github.blog/changelog/20... I like when GitHub adds features like that, to improve the ability of maintainers to keep repos organized.

The GitHub MCP server has *91* tools. ~1/3 of those tools are write ops, so they can create, update, or even delete things on your GitHub account. 😱 I don't always need that-or want it! Fortunately most agentic coding tools + agent frameworks (like Copilot, Langchain) include a way to filter.

Yay AGENTS.md support! Going to rename my copilot-instructions.md files to AGENTS.md today.

"But there was the issue of—like, fucking ICE could be outside [my concert]. And it’s something that we were talking about and very concerned about.” - variety.com/2025/music/n...

Dear developers: please do *not* use LLMs/bots to auto-reply to issues in repositories that you do not maintain. Let the maintainers of the repositories decide if they want to do that themselves! Generated comments from non-maintainers can appear helpful, but are often confusingly authoritative.

www.npr.org/2025/09/08/n... ..welp, now I feel better about about the gopher traps I just setup. (I tried planting only native mint/salvia, smelly plants that gophers don't like, but THEY ATE MY MINT. they went too far)

My kids are now 3 and 6. I feel like I missed watching the 6 year old grow from being a 3 year old herself.. 3, 4, 5, 6. The years flew by as my time was taken by the very demanding second kid. And now my baby's all grown up! 😭

In general: - APIs that are confusing for humans tend to be confusing for LLMs; - APIs that are not flexible enough are more prone to hallucinations; There are so many interesting techniques for using LLMs as user-study participants of sorts. A true LLM-driven design workflow.

LLMs are _really_ good at writing code - not a news at this point. What I find more interesting is LLMs used as beta-testers for new APIs and DSL: - start with idea + docs + examples - LLM-generate lots of code - see what and how it fails. It seems like LLMs tend to fail _similarly_ to humans 1/2

whenever a local small business uses a diffusion model to generate art that has obvious artifacts, I want to volunteer to photoshop it for them to clean it up, but I dont know if that'll offend them. i know theyre doing the best they can with limited funds.

A breathtaking sight! The ultra-rare “Firework Jellyfish” captured on camera nearly 4,000 feet beneath the ocean’s surface. 📹: @nautiluslive.org

I gave a virtual talk today about building agents with Python and LangChain v1 to a great group of students Tunisia. Grab slides here: onedrive.live.com/:p:/g/person... Important point from the slides: not all calls in our systems *should* be agentic! Only introduce LLMs when needed.

Bad Bunny put on two amazing live performances on SNL this year, neither of which are available any more on YouTube - apparently due to music rights: www.reddit.com/r/LiveFromNe... Is there any service I can pay to see those performances again? I crave them, I need them, por favor.

Practicing saying "secede" correctly. You know, just in case.

BREAKING: By an apparent 6–3 vote, the Supreme Court halts an injunction that had prevented immigration agents from racially profiling Latinos in central California. Sotomayor, dissenting, says the decision is "unconscionably irreconcilable with our nation's constitutional guarantees."

Played around with @logfire.pydantic.dev from @pydantic.dev during today's Python+AI office hours. Took just a couple minutes to set up and start seeing traces from agent calls!

I set up evals to test our multimodal RAG flow. Small sample size (10 Qs) but results still interesting: *If we remove img embeddings from search, answer quality still high *If we stop sending imgs to the LLM, relevance & ground truth similarity drop significantly Repo @ github.com/Azure-Sample...

It's been 3 years since my second daughter was born, yet I think I'm still in shock that she's so radically different from my first daughter. Maybe one day I'll come to accept the mysterious ways of gene mixing.

The Pydantic AI team is prepping to release v1, so I spent the day making sure that my Pydantic AI examples are v1 compatible. 🙏🏼 Kudos to the team for fixing the 2 issues I found, within 2 hours of reporting! My upgraded examples: github.com/Azure-Sample... github.com/pamelafox/pe...

Released new version of ndjson-readablestream, for streaming NDJSON (JSON-lines) in the browser: https://www.npmjs.com/package/ndjson-readablestream

👀 commandfor - enabled by default in Firefox 144 Soon, in all browsers, you won't need any JS to open a popover / dialog 🎉 More info in @keithamus.social post here: developer.chrome.com/blog/command...

BREAKING: Gov. JB Pritzker says the Trump administration has sent the Texas National Guard to Great Lakes Naval Base outside Chicago, awaiting orders to deploy to Chicago. Trump administration officials will target Mexican Independence Day (Sept. 16) celebrations. @wttw.bsky.social

Playing around with gpt-oss on my YouTube channel: youtube.com/live/118umy7...

If you're interested in RAG or evals, I recommend following @hamel.bsky.social I subscribed to his email newsletter and am loving the topics: ai.hamel.dev Particularly loved a recent series from his evals course "Stop saying RAG is dead" - hamel.dev/notes/llm/ra...

GitHub Copilot now supports AGENTS.md for agent-specific instructions. Yay for everyone getting along and picking a filename! https://github.blog/changelog/2025-08-28-copilot-coding-agent-now-supports-agents-md-custom-instructions/