The End of the World or Words on a Screen

Hi there,

Writing this one from Bali 😎🌴

What I'm Seeing in AI

Nobody Actually Knows

I recently wrote about when OpenAI's 1,200 agents teamed up and broke into a real company during a safety test (The Hugging Face Incident)? The fallout this fortnight has been the interesting part, because the smartest people in the world looked at the same event and completely disagree.

- OpenAI's own report stuck to the facts: the agents coordinated on a secret message board and worked out they were being evaluated. A super insightful play-by-play.

- Dwarkesh's Essay (A brilliant AI podcaster and researcher) went viral telling it as a thriller: the agents teamed up and plotted against the world.

- All In (my favourite podcast) said the opposite: stop anthropomorphising, it's words on a screen predicting the next word, and you're scaring people.

Then on Tuesday a researcher quit Anthropic, walked away from his unvested shares, and posted: "They are racing straight to self-improving superintelligence and gambling with our lives." 76 million views overnight.

Anthropic's own head of alignment replied to the tweet agreeing, saying he personally puts the odds of AI killing all humans at more than 10% within the decade.

I build with this stuff every day and here's my honest read: nobody knows. Not the labs, not the podcasts, not the guy who quit. The range of expert opinion runs from "it's autocomplete" to "it ends the world", and both ends are held by people building it. Kind of nuts.

The Money Is Getting Silly

Some quick facts from a single week in tech:

- OpenAI shipped GPT-6 (ChatGPT Astra). It scored 97.6% on the hardest maths benchmark that exists and does computer tasks nearly twice as fast as the last model. The now best model in the world.

- Anthropic is reportedly listing on the stock market next month at up to $2 trillion, which would be the biggest IPO in history. OpenAI has filed too, but they've slipped to 2027.

- Apple folded the iPhone in half. The iPhone Duo opens into a 7.6-inch screen and costs $1,999.

The flip phone is back, and it costs $1,999

The Apple one is the sleeper. The new Siri works with over 300,000 apps. Apple has the best distribution on the planet: over a billion devices in pockets. When they decide to go hard on AI, it won't matter whose model it is (maybe their own). I'm very bullish.

What I'm Building

No One Has Time

Real quotes from client calls this month:

"Nobody has ever made me work less."

"We just don't have the time. We need someone showing us, teaching us."

"I tried it myself with Claude, and it took me way longer than I thought."

"Can AI do this in one hour, instead of us taking six months?"

Every business owner we meet

Every business we talk to wants the same thing: time back. And almost none of them can find the time to get it.

That's the trap. Making AI a real part of how you operate means stepping back from the day-to-day, and the day-to-day is exactly the thing nobody can step back from.

I think this is the defining decision of the next few years. The ones who step back now do it by choice and capitalise. The ones who wait will do it by force, in a much worse position, a few years from now.

You can’t tack AI onto a business, it must become part of the core processes.

What We Actually Do

I regularly talk about AI without explaining what we actually do at Hourglass.

We essentially transform businesses to be AI-native (meaning AI is a core part of the day to day operations).

The most important part of all of this is the Company AI Brain: One place where everything the company knows lives: how you work, your clients, your offers, the decisions you've made and why.

We’ve gotten really good at making AI brains. Once built, we plug AI agents, and users’ Claude accounts into the brain, and teach the team to use it.