AI for the rest of us

Welcome to the AI community for everyone.

Hello friends,

Everyone is talking about Moltbook and we are no different. If you haven’t heard about Moltbook, it’s a “social network for AI agents” - a place on the internet where LLM twaddle is the main event. For once Hannah is more opinionated than Charles on the subject. Read on for a good old rant.

Charles is writing this week whilst sitting by the fire and wondering if it will ever stop raining. Perhaps because of the weather he is feeling particularly grumpy. Hannah is also very upset with the weather, mostly because she has the energy of a bouncy puppy who must run around outside at least once a day or she’ll start chewing the furniture. It’s almost time for her walkies!

For those of you based in London our next meetup on Thursday Feb 19th is a bit more technical than our normal agenda. This month we are learning about how to use AI coding tools the right way. Daniel Jones will be speaking about his experience training over 100 developers how to get the most out of these tools.

Have a fabulous week everyone!

Hannah and Charles

What’s Charles reading this week?

Hannah wrote about Moltbot (formally Clawdbot) last week and then ‘Moltbook’ came along as a Reddit-like forum for all of those agents to ‘talk’ to each other. A lot of people seemed to get frightfully excited about this, and not just on LinkedIn. Professor David Reid, a Professor of AI and Spatial Computing at Liverpool Hope University, was on Radio 4’s Today Program on Monday to tell listeners this was “AIs talking to each other, and helping each other for the first time…I wouldn’t say they were conscious and I wouldn’t say they were just prompts either, it is somewhere between the two.”

Professor Reid is a very eminent man, but there’s no reason to assume that the interactions on Moltbook are purely AI. Some posts are coming from humans masquerading as bots, and lots of others are fakes. The whole thing smacks of writing “I’m alive” on a piece of paper, photocopying it, and then saying “OMG look what the photocopier says!”

But whilst LLMs are no more conscious than your photocopier, could a computer built from human stem cells become conscious? And how would we know? My youngest likes to joke that while AI systems require vast amounts of space and horrifying amounts of fossil-fuel based energy, his brain is more capable, fits in a space of about 1,400 cm³, weighs roughly 1.3 kg, and runs fine on water and sandwiches. The broader point is that biological brains operate with power and data efficiencies otherwise unparalleled. The human brain runs on ~12-20 watts. Real data for AI energy consumption remains somewhat elusive, but we can say current AI systems (data centres, GPUs) require gigawatts.

Given biological systems' capability for rapid real-time adaption, might it be beneficial to consider neural systems in the current search for AI? There are two mutually inclusive forms this may take: (1) using biological cells as the substrate for information processing and intelligence; (2) developing new AI algorithms inspired by these biological systems. Dr. Ewelina Kurtys is exploring the first option at FinalSpark, building a computer from human stem cells.

I honestly don’t know how I feel about this yet, but it’s fascinating work and I loved talking to her about it for GOTO Unscripted.

Voice-based interactions are becoming an increasingly important area for Large Language Models. ElevenLabs, a software company that specialises in developing natural-sounding speech synthesis software using deep learning, raised $500 million this week. Writing for Tech Crunch, Rebecca Bellan notes that ElevenLabs’s vision of voice as the mechanism that controls technology is one that is “increasingly shared across the AI industry. OpenAI and Google have both made voice a central focus of their next-generation models, while Apple appears to be quietly building voice-adjacent, always-on technologies through acquisitions like Q.ai. As AI spreads into wearables, cars, and other new hardware, control is becoming less about tapping screens and more about speaking, making voice a key battleground for the next phase of AI development.”

In this newsletter a couple of weeks ago I mentioned how difficult it is in journalism right now, and jokingly noted that one business model open to publishers was being owned by Jeff Bezos. Turns out that isn’t such a great business model after all, with another 300 journalism jobs lost, this time at the Bezos-owned Washington Post. On the technology desk the likes of Pulitzer-shortlisted, cybersecurity legend Joseph Menn are now out of work.

Search traffic at the Post has fallen by half in the last three years, according to its executives. This is undoubtedly partly due to AI summaries, but Bezos's decision to pull a presidential endorsement days before the 2024 election and remake the editorial page to align with his politics is one of a number of such decisions that repulsed its core audience.

The Post lost about $100 million in 2024, which is clearly a lot. On the other hand Bezos earns about $7.99 million per hour so could cover it by sacrificing about 12 hours pay. Bezos paid $250 million for the Post and $500 million for his yacht, Koru.

It is on a much smaller scale but I was also deeply shocked and saddened to see layoffs impact The New Stack following the retirement of their chief editor, the wonderful Heather Joslyn. The New Stack was sold to Insight Partners in 2021. It is a site which I’ve long admired and have freelanced for for many years. Those leaving include founder Alex Williams, veteran tech news reporter Joab Jackson, and editor Vicki Walker, who I was lucky enough to do a lot of writing for and is one of the best in the business. If any of you need good people let me know and I can make an intro.

The UK’s competition authority (the CMA) is trying to do something about Google's AI overlays, with a proposed new set of rules for Google search, including requiring that it let publishers opt out of AI Overview without (as Google presently requires) opting out of search entirely. Google says it’s considering applying this globally, not just in the UK. A lot of publishers have asked for opt-outs, but I’m not sure why. If Google isn’t ingesting your content for AI you don’t have so many bots costing you money, but if consumers are only using (or being given) AI overviews and you opted out, then you lose the traffic anyway. A drum I feel like I’ve been beating for far too long is that the only way to succeed in media is to have an audience that is loyal to you, and not rely on anyone else’s platform whether that is search, social, GenAI or whatever.

Apple, the media narrative suggests, is behind on AI, and that’s perhaps most visible with Xcode, the development environment generally used for building applications for Macs and iThings. Apple has a big developer conference, WWDC, every June where it announces major updates to its tools and OSs, but has rushed a release out early which integrates Xcode with both Anthropic’s Claude Agent and OpenAI’s Codex. The Mac App for the latter was also released this week.

Being behind on AI and, in my opinion, making a horrible mess of the once magnificent MacUI with the Fisher Price My First Operating System™ look of Liquid Glass, doesn’t appear to have dented Apple’s profits, with revenue of $143.8 billion, up 16 percent from a year ago.

There’s a famous story that shortly after Steve Jobs returned to Apple in 1997, the company he helped start in 1976, Dell’s founder and chairman, Michael S. Dell, was asked what might be done to fix Apple, then deeply troubled financially. “What would I do?” Mr. Dell said to an audience of several thousand information technology managers. “I’d shut it down and give the money back to the shareholders.”

Dell’s market cap today: $76 billion.
Apple’s: $3,824 billion.

What's Hannah reading this week?

We’ve reached the singularity! Artificial Intelligence has gained sentience! AGI is here! The world is over (probably!)

Sigh.

Everyone is talking about Moltbook and I’m going to join them!

Although it is incredibly silly, The Moltbook Moment is something we’ll never forget. For a lot of people it’s the first time they’ve experienced AI Agents interacting with each other, and working together and that is HUGE.

It feels magical, sparks our curiosity and helps us imagine what the future might look like. The lightbulb moment, when you realise what this technology is capable of, has happened at viral scale in the space of a couple of days.

I remember how I felt when I first saw a demo of a multi-agent team independently working through a problem - it blew my mind. I went “wow” and then “oh fuck” in the space of 5 minutes as I witnessed a team of agents execute a complex task that would have taken around 6 weeks for a team of humans to complete.

That was early last year while working with the team at Futuria. I also gave a talk about it at QCon in April 2025 where I shared some of my early thinking about how Agentic Automation could reshape work. Discussing The Moltbook Moment this week my friend shared the feeling of whiplash, the speed of progress with this technology is nothing like we’ve ever experienced. It’s bigger than a lightbulb moment, it’s a solar flare!

Witnessing AI Agents create their own religion or share stories about their humans is deeply unsettling until you understand how AI Agents and Moltbook work. A user can tell their Agent “Go and post on Moltbook and create your own religion, it should be consistent with the crustacean theme.” Similarly, although the front page implies otherwise a human user can also post directly to Moltbook, as explained here in mashable. Each Agent on the site is reading posts and those posts are adding to their context about what Moltbook is for and what makes a good post or comment. This context is influencing what each Agent calculates as the “probably right” comment and the “probably best” post. That’s what LLMs do, they figure out the “probably right” answer in each new scenario based on context. This explains why a lot of the comments on Moltbook resembled the AI Slop you see commented all over LinkedIn these days.

It’s amazing. It’s stupid. It’s brilliant. It’s scary.

I’m delighted by the experimental novelty of the whole thing (honestly I am) but I’m also concerned about the risks. Moltbot demonstrates a huge appetite for Agents that can do things. One reason that they’ve been able to achieve it whereas Apple, Anthropic and OpenAI haven’t, however, is because of a reckless attitude to security.

Over 400 malicious skills have been published to Clawdhub which directly install malware on the user’s device. At the beginning of this week the entire Agent database of Moltbook was left wide open to hackers due to an insecure API, giving hackers access to every Agent on the platform via their secret API key. This is serious stuff when you consider that many of the Agents have access to their users' email accounts, and email accounts can be used to unlock a vast number of other applications via password reset workflows.

But the main risk, as I see it, is that people are starting to believe in something that isn’t real. That AI Agents or Digital Workers or Bots or whatever you call them are more than just software. Where does that thinking lead us? How might it harm the people who hold that belief? And who might abuse or manipulate those people for profit?

You can’t put this genie back in the bottle. The Moltbook Moment has communicated what this technology can do today in a way that no start up pitch deck ever could. AI Agents can actually do things. AI Agents can actually work together. The technology isn’t perfect yet but it’s closer than you think. Collectively we need to decide what future we’re building towards if we want it to be more useful than a crustacean-based religion.