Hi there,
Here’s your daily reminder to smile more :)
What I'm Seeing in AI
The Gap
I’ve been thinking a lot about the gap between AI power users, and those that are just scratching the surface. Here’s some stats:
About 49% of US adults have tried an AI chatbot, and only 24% use one every day (Pew, June 2026).
The most common thing people use it for is searching for information: a better Google.
Nearly three quarters of ChatGPT messages have nothing to do with work (OpenAI’s own study).
Companies look much the same. Almost every business says it uses AI somewhere, but only about 10% are running AI agents at scale in any part of the business (McKinsey).
On the flip side, AI agents can now finish tasks that would take a person about 5 hours on their own, and that number doubles roughly every four months (METR).
I think this gap is now so wide it can’t be ignored. Whilst some people are typing word documents by hand, others are running entire companies from their phone texting AI.
Surely this will start being a massive issue??
The Weird Economics of AI
AI models got cheaper again this week. xAI released Grok 4.6 at about half the usual frontier price, and DeepSeek released V4 Pro the same day at about a seventh of that (coverage).
It is a race to the bottom, and I think the subscription era we’re in right now is temporary. Pricing for AI is going to get strange.
My thinking is that the newest, smartest models will stay expensive
Everything below that line gets commoditised to basically zero, mostly by open source. The intelligence you pay for today is free in a year.
On the company side, the figures are also weird.
OpenAI’s IPO paperwork is expected in the next few weeks: a September listing at up to $1 trillion, and the numbers reportedly show them losing about $1.22 for every dollar they earn (reports, 10 Aug).
Nvidia is arranging $500 billion of Wall Street financing so its customers can borrow money to buy its chips (CNBC).
With Anthropic going public estimated around October, the next two months are when the market puts a real price on the AI boom. Until now it’s been up in the air.
What I'm Building
The Software Factory
The next step past chatting with AI has a name: the software factory.
Instead of ChatGPT or Claude answering you one message at a time, a whole fleet of agents runs the job. You tell it what you want. It asks you questions until it understands.
Then you go to bed, and come back to a built app or website.
Humans set the direction and judge the result; the agents do everything in between. BCG wrote a good piece on it this year.
Almost nobody is running factories yet. Even serial founders in San Francisco are mostly just on Claude Code, but it’s cool to look at what’s possible.
Teach a Man to Fish
This connects to a decision we made at Hourglass this week.
Our best clients are the ones who want to learn to build with AI. The hardest engagements are the ones where a business wants everything built for them and stays hands-off.
So it is now an explicit filter in our sales calls: we look for businesses that want to build, not have AI built for them.
If you know someone still using AI like Google, send them my way 😈
Also Happy Birthday Gigi!!
See you next Friday,
Finlay
Explore the rest of my writing here → finlayekins.com/writing
Join our monthly company newsletter here → thehourglass.ai/about