Hi there,
I'm visiting my sister this week in Port Macquarie 😎.
What I'm Seeing in AI
A New AI Model (Sigh)
Anthropic released Claude Fable 5.2 on Monday and of course it’s immediately the most capable AI model of all and can do a million new things.

Source: Epoch AI
For me, however this is the first time a new frontier model has come out and the last one (Opus 5.5) still felt amazing. Maybe this new model is even more amazing. But I have no real drive to switch.
My point in all this is that AI advancement has been so rapid and so ridiculous recently, but now that each model can take on pretty much any work I give it, no one really cares.
Doom on Twitter, normal life at the pub
In AI circles and places like X (twitter) there's constant doom. Every day you read AI is about to take every job, super-intelligence is a year or two away, and we're all about to be overthrown.
Most charts like the following predict fully automated coding by early 2027 and super-intelligence (ASI) by December 2027.

On the other hand, you go to the pub, or step into the real world, and no one talks about this stuff at all, people are barely even using computers correctly.
So what's the truth? Are we all about to get overthrown by AI, or is it just business as usual?
Personally, I believe 99% of people will just keep living their life, without realising much has changed. We already use AI in most uni and office work. In 10 years we might go to the GP and it's run by AI. In 20, AI might be doing the dishes. It'll creep in, and most people won't really notice, let alone will they be enslaved by it. We evolved to fear new things, and there’s always been hysteria, but so far (fingers crossed), we’ve always been wrong.
P.S. crazy to see Nvidia's CEO Jensen Huang admitting to multiple affairs this week. Not often you see that from the head of the world’s biggest company.
I had to call Jensen and tell him we’re no longer friends
What I'm Building
Security, security, security
Now that we're established and have a good set of products, we're going really deep on security and systems, and cleaning up how we run everything.
The biggest piece is the brain. Every client gets one: an AI that reads their Slack, email and meeting notes so it knows how the business actually runs. That's a lot of sensitive data.
So the brains run on open models, with zero data retention: nothing is stored and nothing is used for training. The brain now refuses any model call that can't guarantee that, and the list of approved models is being hard-coded, so nobody, human or AI, can slip in one that keeps data.

Why so strict? One client's contract says that if we ever part ways, we have to return or destroy all their data within 30 days. You can't promise that if a model provider somewhere is keeping a copy.
Help us pick our first merch
On a lighter note, I designed some Hourglass merch: tees, hoodies, caps, a notebook and some stickers. The team has narrowed it down to three designs, help us pick one!

1. Human in the loop

2. In good company

3. Pixel pair
Which design should we print?
I'll share the winner next week.
Is AI going to kill us, or is it just hype? Hit reply, I looove replies.
And if someone you know would like this, forward it on.
See you next Friday,
Finlay
Explore the rest of my writing here → finlayekins.com/writing
Get Open Sauce, our weekly Hourglass newsletter → thehourglass.ai/open-sauce

