The AI Work Factory

Hi there,

“F@#k, this is f@#king blowing my mind." - A CEO seeing our AI setup yesterday

What I'm Seeing in AI

Anthropic is going public

On Monday Anthropic filed to list on the Nasdaq under the ticker "ANT". The filing describes a company that's never really existed before:

$6.6M revenue per employee. That's more than Nvidia, Apple, Google or Meta ever. The one company that beats it is OnlyFans 😆
Customers spend about 9x more after their first year. Good software companies brag about 1.2x.
$41M in revenue three years ago, $16.3B in the first six months of this year.

Epoch AI chart: revenue per employee, Anthropic and OpenAI above Nvidia, Meta, Apple and Google

Revenue per employee (Epoch AI's estimate from May; the S-1 puts Anthropic at $6.6M)

Anthropic says the gap between what Claude can do and how widely it's used is one of its biggest growth opportunities. The models are ahead of most businesses, and someone has to close that gap, that’s us baby.

How AI gets sold next

Most AI today is sold like Netflix: a monthly subscription per person, whether you use it or not. This doesn’t last longterm.

On Wednesday Anthropic gave big companies a way to see what their Claude spend is buying. Claude reads a sample of a team's chats, groups the work into workstreams, and shows the spend, what got made and where people got stuck for each one.

That only matters if pricing is heading somewhere else. Dario Amodei, Anthropic's CEO, said on a podcast: "At some point we're going to see 'pay for results' in some form."

If you're an AI company that only sells subscriptions, you're cooked. The money follows the outcome, and clearly based on the S-1, companies are finding lots of good outcomes.

The example Dario gives: When AI tells someone to restart their Mac, those words are worth a few cents. When it tells a drug company how to redesign the molecule they're developing, the same amount of AI could be worth tens of millions of dollars.

What I'm Building

The work factory

People in AI talk about a "software factory": agents that take an idea and keep pumping out code. I've spent the past few weeks building something like that, except it runs the whole business, not just the code. I call it a work factory.

The factory picks up my meetings, emails and Slack, and if something needs doing, they go and do it. Any new deal, meeting or action item goes through the factory, and because the brain and connectors know everything about the business, it knows what to do.

All that reaches me is one decision board: do you want X or Y? My emails get drafted, I read them and hit send. It's shipping about 100 changes a week.

Finlay's factory page: six stages from Factory operational to Utopia, and the line from work context through triage, build, ship and monitor

My Factory Page

I think this is where the future of work lies. The AI is always running, looping through the work, and humans are left with the decisions it can't make and the fun stuff: the meetings, conversations, and creativity. Plus the odd "can you turn the power off downstairs" until robots are around.

The intelligence is already good enough. What matters now is that it sees everything, so we record everything, log everything, and put it all through one system of record.

Open Sauce gets videos

Open Sauce is where we give away the skills and playbooks we use at Hourglass, for free. Matt rebuilt it this week so everything sits on one page.

The newest one is a skill Matt built for himself. You give it a recorded client interview and it hands back a finished YouTube case study. It researches the client first so every number on screen traces back to something they said, whacks on the transcript, cleans up the audio, builds a Diary of a CEO style cold open, and checks every render for sound, lip sync and captions before anyone sees it.

The first one out is Damien, CFO of Heaps Normal, who built his own credit app:

And an earlier one, with our favourite recruiter client, Tom:

The videos look sick, and I can’t believe how fast he can pump them out