Jason Birchall

TIL: I don’t have to stick my head out of the car window to know which pump to use. 😅

Thanks for the read, Lorin. I’d be keen to hear your views on how we can use the SRK model to improve observability/dashboards. I guess it's a case of making the “right” observation, the path of least resistance.

🎉 Congratulations to the Ministry of Justice Splink Team (@robinlinacre.bsky.social) for securing the win in the Open Data category, sponsored by @odihq.bsky.social at the OpenUK Awards 2025!
#openukawards #opensource #opendata

As 2024 ends, my time at the Ministry of Justice also concludes. Over seven years, I've worked with amazing people and learned a lot. I'm proud of our progress and excited to take on new challenges in 2025 as I lead the SRE effort for the GOV.UK project at GDS. #sre

It's the same with TDD, in that thinking about tests up front is a design tool that shapes how you think about your systems before writing that first conditional statement. The value is in the thought process, not the artefact it produces.

I think that gets lost when we automate these away entirely. We lose the frameworks that help us understand our systems

Thanks for writing this
While the thought of automated incident response is super appealing, especially during those 3am pages, it's important to remember that processes like incident response, specifically retrospectives like postmortems, are intended to facilitate human thinking and learning.

I couldn't fit it all in. But I'm excited. It's the Pleias 1.0 models.

They Said It Couldn’t Be Done A Blog post by Pierre-Carl Langlais on Hugging Face

Okay, this feels like progress!

“These represent the first ever models trained exclusively on open data, meaning data that are either non-copyrighted or are published under a permissible license."

huggingface.co/blog/Pclangl...

A map of a neural network. On the left are five trolleys, linked to a series of circles representing the neural network, linked with a dense set of lines indicating possible connections. On the right, the result of the neural network is a number of people tied to train tracks.

Dusting off my best work

What if it could test its guesses? Clone the production environment, confirm the issue, try a rollback or fix forward, and validate automatically.

I get the scepticism around AI tooling in #SRE, but Meta’s approach to incident response is interesting: 42% accuracy in identifying root causes. That’s pretty good for large-scale systems.

Wouldn’t it be cool if it went further?
https://tinyurl.com/fh4tpx8c

Screwdrivers are going to make hammers obsolete. Silly right? That's what most tech discussions turn into.

Looks good 👀
Thanks for the suggestion. I've never had a problem with Pipenv, but I'll take a look.

I completely agree with you, Joel. I also really like the idea of starter packs. It helps newbies get set up quickly and follow communities that have been built up over many years on the platform we dare not speak. Mastodon may seem like a scary jump-off point.

Bluesky or Mastodon, that is the question.

Still getting over the fact that a lunatic can stand up in front of a crowd and scream about migrants eating cats and dogs and it gets him elected. Maybe I'll never get over it.

“Education isn't something you finish.”

~ Issac Asimov

"The easy path today makes a hard path tomorrow. The hard path today makes an easier path tomorrow."

A Tiny Thought worth reading in the Farnam Street blog.
fs.blog/brain-food/o...

“The master (of chess) does not search for combinations. He creates the conditions that make it possible”. Ivan Chernev, 1957

… and change one variable to make it look original 😝

Exploring Blue Sky feels like a breath of fresh air—a reminder of the excitement I had when I joined Twitter years ago.

No platform is perfect, but it feels cleaner and more focused. It’s early days, but I’m genuinely enjoying this space's vibe and potential.