AI for the rest of us

Welcome to the AI community for everyone.

Hello Friends,

It’s been a momentous week. Hannah experienced her first earthquake in Sendai, Japan and this newsletter had its first week of 1000+ readers (and rather more subscribers). Thank you all for being here.

We also have momentous news about our next conference!

We’re excited to announce a new one day conference: Agent Craft. Agent Craft is a gathering for Agent builders, a space where we can share learning and connect as a community. Hosted in partnership with The AI Summit London and Brainstation, we’ll be wrapping up London Tech Week in style on Friday 12th June in the heart of Shoreditch, London.

Agent Craft is an unconference, this means it’s the people in the room that drive the agenda. We kick off the day with a few select keynotes and demos and then it’s down to you. What do you want to talk about? What problems do you need help with? Pitch your idea for a session and connect with the people who care about the things you care about.

Newsletter readers can get 50% off tickets for the next 2 weeks using code AI4NEWS. With just 100 tickets available don’t miss out!

For anyone who has attended AI for the rest of us conferences in the past, don’t worry. The 2 day, multi-track format is not going away. It’s just moving to February!

In other news, the audio version of the GOTO podcast we both did to talk about Charles’ book "Kubernetes at the Edge” is out, so if you prefer your podcasts in audio rather than video format give it a listen.

Have a momentous week!

Hannah & Charles

The AI Native shift is happening now — download the free eBook or get left behind

What’s Charles reading this week?

I was shocked and saddened to learn about Kent Beck’s Parkinson's diagnosis this week. His framing of it as the “time value of time” is one of those very Kent phrases that sticks in your head.

Kent has dedicated his life to, as he puts it, trying to make geeks feel safe. I’m a huge admirer of his work, and had the pleasure of spending some time with him at YOW! last year. I loved his keynote (the video isn’t out yet but I wrote about it here) and he was, in person, every bit as kind and delightful as you would expect.

As he says, “If you were thinking of booking me for coaching, consulting, or a talk for your team, might I suggest you contact my business manager sooner rather than later?” I would very much encourage you to do so.

I find media narratives — the framing that news outlets, journalists, and media organisations construct around a particular event, issue, or person — fascinating. I’ve only read some of the research on them, but a short summary would be that the brain imposes story structure on experience almost automatically, turning sequences of events into causes and effects, heroes and obstacles, beginnings and ends.

The cognitive scientist and artificial intelligence theorist Roger Schank argued that memory itself is essentially stored as stories — we remember experiences as narratives, not data.

Organisational theorist Karl Weick used the term sensemaking to describe how humans construct meaning from ambiguous events. We grab available narrative frames, traditionally supplied by the media, to make sense of things we don't fully understand. Media narratives are powerful because they exploit something deep and automatic in how we process the world. The media isn't creating a vulnerability so much as plugging into existing cognitive architecture.

Controlling the narrative helps explain why someone like Musk is pouring money into Grok. But, whereas in conventional media influence pushes a group towards an accepted view, chatbots can be rather more personal, with alarming consequences.

The Guardian’s Edward Helmore reports, for example, that Florida is set to open a criminal investigation into OpenAI over ChatGPT’s influence on an alleged mass shooter. The State attorney general said the inquiry will look into whether the AI tool “offered ‘significant advice’ to [the] campus shooting suspect”. On 404 media Samantha Cole shares an alarming set of experiments where researchers cosplayed as a vulnerable user in various scenarios, like romance-seeking or delusions of grandeur, and watched how chatbots responded over the course of 100+ turns. You’ll be unsurprised as to which chatbot was the worst.

Along with the LLMs themselves, Musk and the other AI vendors also spend a lot of time and money trying to shape the media narrative around AI. With this in mind, I recommend taking the time to read this terrific essay and Decoder podcast from Nilay Patel, chief editor at The Verge, in which he explores the disconnect between the software industry’s view of AI and the vociferous public opposition to it, allied to deeply pessimistic expectations about what it’s going to do. The essay is US centric, but Patel is a deep thinker, and it really is worth your time.

You can see a narrative battle playing out in real time with Anthropic’s Mythos. There are basically two schools of thought. One says: “Anthropic is using doom mongering as a marketing strategy, and the reason they’re holding Mythos back has nothing to do with security, and everything to do with capacity constraint”. OpenAI has been pushing this line, going on podcasts saying that the limited Mythos release is simply "fear-based marketing", which is, um, ironic. Pot…Kettle on line 2 for you. The other argument says: “Mythos really is a big deal for security.”

Narrative hates nuance but both things are likely true. Anthropic is undoubtedly capacity constrained, and likely couldn’t roll out Mythos quickly. The untenable demand for Claude Code sees Anthropic exploring new approaches to rationing its service, including experimenting with removing Claude Code from the Pro plan, causing much wailing on Reddit and, I’m told, on X.

Capacity constraints also partly explain the circular financing which underpins AI. If you are selling your service at a loss, you can’t invest to scale it. Ars Technica’s Jeremy Hsu reports that Amazon has invested another $5 billion in Anthropic, which Anthropic can use to purchase more AI chips and cloud computing resources from Amazon. If and when the AI bubble finally bursts, circular financing is one of those practices that will look very, very dodgy indeed and might even be made illegal.

Equally, Mythos clearly is very capable: Mozilla’s CTO has said that Mythos found 271 security vulnerabilities in Firefox 150, and is “every bit as capable” as world’s best security researchers.

At some point, the narrative will likely collapse to one or the other, although social media seems to have extended the lifespan of preferred narratives significantly, whilst also making it much harder for any single shared narrative to emerge.

The big tech story this week is that Tim Cook is stepping down as CEO of Apple. It is a very Tim Cook transition, well thought through and carefully signposted, so there was no real impact on Apple’s stock price, although I’d expect some churn at the top over the next few months. I thought Cook’s letter announcing it was one of the best things he has written since his “coming out” essay for Businessweek.

There is a firmly established narrative around Cook’s time at Apple which says “he’s no product visionary and it's all about profits”. It irritates me. Partly because it is unfair, and also because Cook, by all accounts, would be the first to tell us he doesn’t want to be judged by the numbers alone. Or as he put it himself at a shareholders’ meeting early in his reign, “When we work on making our devices accessible by the blind, I don’t consider the bloody ROI”.

He has been, by any objective measure, an outstanding CEO. During his time Apple launched the Apple Watch, as well as AirPods, both of which are undeniable hits. He has also built a services business that now accounts for approximately 21% of Apple’s revenue — it wasn’t so long ago that the narrative was Apple couldn’t do services.

He’ll leave the Mac in the best state it's been in for years, including the recent launch of the Neo, having shifted from Intel to Apple’s ARM-based Silicon. Designing the M-series chips in-house was a huge engineering undertaking and the performance-per-watt results have been extraordinary. One way you can see how much more the Mac matters than it used to is that Google has now joined Anthropic and OpenAI with a desktop Mac app for Gemini.

Cook has pushed Apple deeper into health technology than Jobs ever did. The health sensors in Apple Watch, ResearchKit, and CareKit, along with the hearing health features added to AirPods, represent a coherent product vision not just margin optimisation. Likewise, he has made environmental commitments a much more central part of Apple's identity than they were under Jobs, with serious commitments around carbon neutrality, renewable energy, materials, and packaging.

There have been misses of course. Apple’s once glorious UX has suffered, particularly on the Mac, from what John Gruber brilliantly termed a bad Dye job. The VisionPro is technically impressive but not commercially successful. And then there is the question of AI. There is nuance here — some of my favorite iOS features are AI powered, and Apple has been building features under the Coding Intelligence moniker into its Xcode IDE via third party integrations.

But the LLM-based Siri has been promised and talked up but never delivered, and The Information reports that Apple has sent several hundred developers on the Siri team on a ‘bootcamp’ to learn how to code with AI.

Away from information theory, The BBC’s Kali Hays reports that Alan Dye’s new employer Meta (hat tip to loyal reader, err, my wife for sharing this story) is tracking employees' keystrokes and mouse clicks through a new internal tool called the Model Capability Initiative (MCI), and using the data to train its AI models. One employee described this as "very dystopian".

Just how dystopian do things have to be for a Meta employee to find them so? Well the company has already cut 2,000 jobs this year and job listings have plummeted from 800 to just 7. CEO Mark Zuckerberg (or perhaps his AI replacement) has pledged to spend around $140 billion on AI in 2026, nearly double last year, as part of an aggressive push to position Meta at the forefront of AI. So very, I’d say. As noted last week that push is producing better models though.

As this week’s newsletter has been quite meta on media I’ll end by saying that Ken Fisher, Founder and Editor-in-Chief of Ars Technica, has published an updated AI policy. In an article announcing it he says, “Our approach comes from two convictions: that AI cannot replace human insight, creativity, and ingenuity, and that these tools, used well, can help professionals do better work”. This is something everyone in the media is grappling with, and I’m pleased to see these kinds of statements being shared in public.

Hannah and I feel similarly. We write this newsletter the old fashioned way — i.e. we read the news, and then sit down at a keyboard and type. Writing it is the highlight of my week, and we’re genuinely thrilled you read us!

What's Hannah reading this week?

Last night at dinner in Sapporo we ordered our food and drink using a tablet and the food was delivered by a robot waitress. As I walked into a department store I was greeted by an impossibly cute penguin robot gazing up at me, chirping and following me around (a high end toy for young children and forty year old nerds alike!)

It feels a lot more “normal” to encounter a robot here in Japan than it does at home in Yorkshire and I’ve been wondering why. Is there a cultural barrier? Infrastructure? Cost? I expect like many things, the future is already here, it's just not evenly distributed. We’ll have robots in York one day.

This week a humanoid robot won the Beijing half marathon. With a time of 50 min 26 seconds the robot smashed the human half marathon record. That’s not really the interesting part though. This was the second year that robots ran alongside humans in the race and the performance of said robots has dramatically improved since this time last year. A very visible and relatable example of how much AI has progressed in the past 12 months.

20 robots competed in 2025, however most did not finish the race. More than 300 robots from over 100 teams competed this year with most of them finishing the course
In 2025 most of the robots were operated by remote control, in 2026 38% of the robots ran the race autonomously

If you haven’t checked out the robots that competed then go and look that up now - here's the BBCs coverage, and a compilation of the cute little guys that did not win! They are brilliant, hilarious and creepy all at the same time.

This dancing robot had to be restrained for being too exuberant. I can relate.

I approve of robots that bring me food, I approve of robots that run fast (as long as they aren’t chasing me) and I definitely approve of robots designed to entertain me. If you’re an elder millennial like me then you’ll remember Robot Wars on the TV, presented by Craig Charles. I think it’s time to bring back Robot Wars! Imagine what they could do today!