Hi there,
Last week I said AI had gone quiet, but not to get complacent, because things move fast.
Well, this week was the biggest in AI this year:
Two IPOs, one all-powerful intelligence, and one redefining shift in how we use AI.
The AI Cold War Goes Public
Here’s the timeline:
Ten days ago, Anthropic (the makers of Claude) confidentially filed to go public
Exactly one week later, OpenAI did the same
The two biggest AI labs are now racing each other to the stock market, the same way they’ve raced each other on everything else this year: models, products, talent.
The numbers are incomparable to any company in the past:
Anthropic was valued at about $965 billion at its last raise. The underdog has, on paper, overtaken OpenAI ($852 billion in March)
Both are expected to list at over $1 trillion
Together, that’s roughly the market cap of every company on the ASX combined 😳
The timing is less petty feud and more cold war. Each lab was waiting on the other to strike first, because the moment one filed, the other had to follow.
OpenAI alone projects burning $115 billion through 2029. Either way, whoever lists first soaks up the scarce capital.
What I’m Seeing in AI
As if that wasn’t enough, Anthropic also released Claude Fable 5, the most interesting model release in years.
Backstory
Anthropic built a model they decided was too powerful for public use (Mythos), and reports had been building about it finding security holes in major companies’ systems.
The full version is only available to ~150 vetted organisations through Project Glasswing: they report over 10,000 cyber vulnerabilities found so far.
The rest of the world gets Fable 5: the same model, with some guardrails and a high price tag
After a few days using it, I personally believe we have reached AGI.
AGI (artificial general intelligence) is what these labs were founded to build: think an all-knowing digital intelligence, AI that can do basically any thinking work a human can, across the board, not just one narrow task.
OpenAI’s founding charter is literally about achieving it. Everything else, the chatbots and the coding tools, is the journey; AGI is the destination (AGI on Wikipedia).
Fable 5 is the first model that feels like a thought partner, not a tool I’m operating. It adds judgement, feedback and taste. It feels like it knows more than me.
No one can agree on the definition of AGI, but I truly believe we’ll look back on this week and think “that’s where it started”.
One of AI’s top dogs, Andrej Karpathy, called it: “a major-version-bump-deserving step change forward”.
What I’m Building
Last week I mentioned the X thread where I shared our company’s AI skills. I’ve now turned it into a public open-source repository: hgstack.
It’s a collection of the AI skills, rules and workflows we run our company on: ready-made instructions that turn an AI assistant into something that does real, reliable work.
The most interesting part of building it was the installer. It reads how your setup is configured and merges around it rather than overwriting. Just because something works for us doesn’t mean it works for you, and generalising our setup to anyone’s machine turned out to be the hardest problem to solve.
It also reflects something bigger to me: the future of software is personalised. Everyone ends up with their own CRM, their own tools, their own AI setup, built around how they work.
Off-the-shelf software will adapt to you, not the other way around. The deeper theory on this is written by Garry Tan (president of Y Combinator): thin harness, fat skills.
Stop Prompting, Start Looping
One more shift from this week. Peter Steinberger, a well-known developer and the maker of OpenClaw, put it bluntly:
8.1M views…
It’s been coined Loop Engineering.
Instead of prompting AI and checking each answer, you build a loop that runs it, checks the result, fixes what’s wrong, and goes again, while you do something else. All you need is a really good definition of “done”: a set of tests the AI has to pass, and it works continuously until it gets there.
Anthropic published a research piece on the same idea: models improving the systems that improve the models. Crazy (and slightly scary) read on where things are going.
Broadly, AI is becoming more and more autonomous, to the point where it’s almost improving itself.
So that’s the week: AI is now reaching financial markets, coding, and white-collar work all at once.
Interested in your thoughts, especially if you’ve tried Fable 5.
And if someone in your world would get something out of this, please share it.
See you next Friday,
Finlay
Explore the rest of the newsletters at finlayekins.com/writing.