AI for the rest of us

Welcome to the AI community for everyone.

Hello friends,

Charles is on holiday this week so we have a special guest writer contributing to this week’s newsletter. Matt Saunders will be known to many of you as the co-host of our London meetups but did you know he’s also Lead Editor for DevOps News at InfoQ? And the host of the popular community London DevOps? And VP of DevOps at Adaptavist? And a dab-hand at home-made pizzas too!

Hannah is off to Wales for a family holiday and has been diligently preparing some stinky bait for a crabbing expedition. There is a tupperware container of bacon that has been strategically left in the sunroom for a week. She’s actually scared to open it now!

This week Hannah was in London looking at venues for the next AI for the rest of us conference in February 2027 and we think we’ve found the perfect place! More details to come in the coming weeks including super-early-bird tickets and a call for proposals (CFP).

Have a wonderful week!

Hannah & Matt

What’s Hannah reading this week?

This video of an AI interviewer made me laugh. I hope I never have to go through an AI interview because now I’ll be too tempted to call myself Chick Bongo and ask for two plumes of gauze and a raccoon protocol. AI is writing CVs, screening CVs and interviewing candidates … maybe it’s only a matter of time before AI hires itself and cuts us mere humans out of the loop entirely.

DeepSeek’s agent harness, released as an open-source project just over two weeks ago has already gained over 200k stars on GitHub. The agent harness is centred on the idea that everything is a plugin, making it versatile and easy to extend. Sentiment in the developer community is that it’s “insanely good”. Unlike proprietary harnesses this one is being built in the open and will certainly benefit from the momentum a thriving community of contributors can create. Will this become the Kubernetes of agent harnesses? We will see!

This week started with a mystery, a new model called Ox Alpha was launched in “stealth mode” with none of the frontier labs claiming the model. The open model has impressive benchmark scores, especially for coding tasks. The anonymous launch created some buzz, with a lot of speculation about who the creators might be, with the mystery being solved on Thursday when Z.ai announced Ox Alpha was indeed a preview of their new GLM model.

100 of the world’s leading technology companies have co-signed an open letter stating that time is running out for cyber security. With current advances in AI cyber-capability the group urges the world to take this threat more seriously and act with more urgency. “We have a limited window to strengthen cyber defences.”

The letter outlines the new challenges and provides 3 principles for response.

Recognise that status quo security won’t be enough
Empower more defenders with cyber-capable AI
Mobilize a collective response

I have been working at the intersection of cybersecurity and platform engineering for years now and I see how overwhelmed most security teams are. For many I’m sure the message to “do more and do it faster” simply isn’t an option. Automation, whether that be via coding agents or more traditional methods, has to be part of the solution. We need to identify and remove the human-speed parts of our processes and start working at machine-speed. That’s what attackers are doing. If I were a security leader today I would be standing up a platform engineering team specifically for security.

Although not exactly AI news I was happy to see that Twitter could be making a comeback. I miss Twitter and I would love to see a new home for the tech community on the internet. It’s not X, it’s not Bluesky, it’s not LinkedIn. There really isn’t a place where we can find our people in the way that Twitter used to. Operation Bluebird is counting on a court ruling to allow them to use the name: “X seemed to have given up intellectual property claims over the word “tweet” and the Twitter bird logo—and possibly even on the word “Twitter” as well.” The platform isn’t ready yet but the team behind the rescue mission are asking for donations and investment to make it a reality.

What's Matt reading this week?

Another of those big money AI acquisitions came along this week; with Nvidia seemingly about to part with billions to buy Hugging Face, which is home to all manner of smaller, more specialised AI models, many of which we can run on our own kit at home. The AI world at large is dominated by the centralised frontier models from OpenAI, Anthropic and co, but I’ve always tinkered on the fringes of technology. I’ve been getting huge value from downloading and running models locally, with terms like quantization and terminology like PyTorch, Transformers and LoRAs and GGUFs keeping me entertained long into the night. Hugging Face has long been a home for this sort of thing, and my first reaction to them being acquired was that of it being a threat to our ability to use an open-source-like bazaar model of innovation with AI. It’s much like we worried Microsoft would ruin GitHub during that acquisition; a threat which honestly I don’t think really came to pass. So long live Hugging Face – we need Nvidia not to ruin it to help keep AI capabilities decentralised.

Data centres are big news this week, with approvals for anonymous grey water-guzzling power-hungry data centres in places as disparate as Essex, Slough and Lanarkshire making the news. The Green Party is calling for a halt to planning approvals, with the backdrop being a controversial approval in Brick Lane in East London. Data centres aren’t new; because I’m ancient, I first worked in one more than a quarter of a century ago. I was involved in a project to convert a derelict textile factory into a data centre and offices, literally at the bottom end of Brick Lane. It’s still there now, and there’s also been a data centre literally on the Truman Brewery site since then too. They’ve been springing up all over the place ever since, and have been essential to modern life since we first started sending emails and doing research on the Internet. But that’s no defence, it feels like there’s a tipping point near, at least in public perception, and stricter regulation isn’t far away. I hope it leads to us all questioning our “all you can eat” approach to AI.