Stop Optimizing For Bible Printing
For its first fifty years, the European printing press mostly made things that looked like manuscripts. Printers cut type in the same hands the scribes had used. They copied the page layouts. They printed what scribes had been copying: Latin Bibles, liturgical books, indulgences, legal forms, schoolbook grammars. The business model was to replace the scribe. Nobody was trying to change what a book was.
The forms we now think of as native to print took longer. Luther's pamphlets come in the 1520s, seventy years after Gutenberg. The first scheduled news periodicals, the Dutch and German corantos, don't appear until 1620. By the time the press had produced its most distinctive output, it was two centuries old.
I think we're somewhere around year fifty with AI.
The obvious thing to do with an agent that runs forever and barely costs anything is to pay it to do the jobs humans already do. That's what most of the AI businesses being built right now are: "AI for X," "AI agencies," "AI SDRs," "AI services." These businesses work. Some of them are printing money. I am not telling you they don't work.
But they are selling bibles.
I know because I built one. For eight weeks earlier this year I ran, solo, a German real-estate lead-generation business in a tax-optimized property niche. Interactive sites with calculators and quizzes, an educational hub, dozens of geo-targeted city pages, a partner referral portal, the usual LLM-flavored content factory underneath. On top of that various proactive agents generating inbound traffic, agent automatically doing SEO & GEO, agent coordinating publishing across YouTube, Reddit, Facebook, Emails and Websites. Cost a few hundred euros a month to run, including the model bill. Sourced enough sales for me to take five figures in revenue-share over those weeks. As businesses go, it ran.
The whole time I was building it, I told myself I was doing something agentic. I was using LLMs natively. I was orchestrating. It was agentic.
It was, in fact, not agentic.
What it was, looked at clearly, was a perfectly ordinary B2C marketing operation with the people swapped out. The org chart anyone in the field would draw was the org chart I had: content team, SEO team, outreach team, sales ops. I'd just replaced each role with an agent that did roughly what the role would have done, slightly worse, for a tiny fraction of the cost. The agents barely spoke to each other. They did their job and dropped a file off for the next one. Score this article. Mail this lead. Cut this video. It was a relay race in agent costumes.
What I had built was a productivity gain over the human version. X percent more productive at cost than humans. That is not nothing. But it is the gain you get from making the existing process cheaper, not from doing a different process.
The shape of the work was identical. I had, in retrospect, organized the technology around the process, not the process around the technology.
This sounds like a cute distinction, but it is everything.
The 10x and 100x gains, the kind that make a technology generation-defining, do not come from getting slightly more efficient at the same thing. They come from doing a thing the previous regime structurally could not do. The press's first fifty years were dominated by Bibles. What broke European life open was a six-page pamphlet in vernacular German that traveled across the continent in a year. The press could do that. The scribes structurally could not. The pamphlet is the form. The Bible is the substrate.
The question worth asking, the only one worth asking if you are trying to do something that matters, is: what does it look like to organize the process around the technology, instead of jamming the technology into the existing process. That's the lens. That's what we should be looking through.
I have suspicions. I think it has to do with the swarm being able to communicate natively. I think it has something to do with radically new forms of information flow. I think it has something to do with sheafs. But these are just guesses. The actual native form may look nothing like that. We will probably know it when we see it, and not before.
The honest, mildly terrifying thing about this is that the way you find the new form is by trying to build it before you understand it. The printer who eventually printed Luther's pamphlets did not start by saying "I would like to invent a new genre of vernacular religious-political broadside." He started by needing short sale cycle recurring revenue in between big book prints. The shape was found by thousands of accidents and people willing to be wrong about it over and over again.
So I've walked away from the agency. It worked, and I don't want to spend the next five years getting really good at making it work better.
I might be wrong about all of this. Year-fifty printing-press analogies are exactly the kind of thing one writes confidently and looks foolish about ten years later. But the thing I'm sure of is that I was bored at month two of running a business that worked, because at some level I could feel that the technology was capable of something I had not yet imagined how to use it for. That feeling, more than any analysis, is why I'm doing the next thing.
If you find yourself excited by an AI business that is essentially "the existing thing, but cheaper" (and to be clear, those will continue to print money in the short term), I think it's worth at least asking whether you're being served Bibles. The Bibles are real. The demand is real. The margins might even be real. They're just not what the press is for and over time you will get annihilated.
I want to be one of the people figuring out what the equivalent of pamphlets and corantos turns out to be, willing to be wrong about it often, in public, until something catches :)