A $50,000 Machine for a Job Your Browser Now Does for Free
On this day in coding history
On May 29, 1985, Eastman Kodak introduced something called the Ektaprint Electronic Publishing System. It was a turnkey box for editing, updating, and printing documents with both text and graphics, and it cost fifty thousand dollars. The publishing software came from a startup called Interleaf. The hardware was a Sun Microsystems workstation. Kodak wrote the front-end glue that tied it together and added its own scanners and printers. You bought the whole thing as one unit, plugged it in, and your company could finally produce manuals and technical documents without a print shop.
It is easy to read that and shrug. A document editor is not exactly the moon landing. But sit with the price for a second. Fifty thousand dollars in 1985 is well over a hundred and forty thousand in today’s money. That bought you roughly what a free tab in your browser does this afternoon: type words, drop in a picture, version the result, send it to print. The interesting thing about the history of programming is not the famous milestones. It is how reliably the expensive, specialized, locked-down thing becomes the cheap, general, invisible thing. KEEPS is a small and almost forgotten data point in a very long trend.
The shape of the trend
Software has been on a hundred-year march from scarce to ambient. Ada Lovelace described an algorithm for a machine that was never built. For a century, programming was a thing a handful of people did on a handful of machines, if they could get time on one at all. Grace Hopper spent the 1950s arguing that programs should be written in something resembling English, because the people who needed computers were not mathematicians and had no patience for symbols. She was told it would not work. FLOW-MATIC, and then COBOL, proved it would. Each of these moves did the same thing: it took a capability that lived behind a wall of expertise or cost and moved it closer to whoever needed it.
KEEPS sits in the middle of that story. The capability was “lay out a professional document.” In 1985 that capability lived inside a fifty-thousand-dollar appliance. Within a few years, desktop publishing on a personal computer ate the low end. Within twenty, the web ate the rest. The Sun workstation under the hood of KEEPS was itself a step in the same direction, a machine that put real computing power on one person’s desk instead of in a shared mainframe room. Interleaf reportedly drove a meaningful chunk of Sun’s early workstation sales, which is its own quiet lesson: the software that makes a platform worth buying often matters more than the platform.
Why the price keeps falling
The mechanism behind this is not magic, and it is worth naming because it is still running. Three things compound. Hardware gets cheaper per unit of work, relentlessly. Software gets abstracted, so the hard parts get written once and reused by everyone who comes after. And distribution gets cheaper, from shipping physical turnkey boxes, to selling shrink-wrapped disks, to a URL anyone can open.
Each layer stands on the one below. The browser-based document editor you use for free is sitting on decades of solved problems: rendering, fonts, undo, collaboration, storage. Nobody using it has to think about any of that, which is exactly the point. The work did not disappear. It got absorbed into the floor.
Where this leaves us, writing code in 2026
I think about KEEPS because the same compression is happening right now to programming itself, and it is happening fast enough to feel strange.
For most of computing history, writing software was the scarce, expensive, specialized capability, the thing behind the wall. You needed years of training to turn an idea into working code. That is the wall that is currently moving. Describing what you want in plain language and getting working code back is starting to feel like what desktop publishing felt like to someone who used to send everything to a print shop. The capability is sliding from “requires an expert and a budget” toward “available to anyone who can describe the problem.”
This is the part where people usually pick a side and either celebrate or panic. I would push back on both. When document layout got cheap, we did not end up with fewer documents or worse ones. We got vastly more of them, and a new set of skills became valuable: knowing what was worth saying, designing for a reader, judgment about what to make in the first place. The grunt work collapsed and the taste work expanded to fill the space.
I expect the same shape here. As the mechanical act of producing code gets cheaper, the scarce thing becomes everything around it. Knowing which problem is worth solving. Understanding the domain well enough to tell a plausible answer from a correct one. In my own corner of this, applying machine learning to biology, that gap is enormous. A model will happily generate a confident analysis of a dataset it has fundamentally misunderstood. The code is no longer the hard part. Knowing whether the result means anything is the hard part, and that has not gotten cheaper at all.
So the fifty-thousand-dollar machine from May 29, 1985 is not really a relic. It is a small picture of where everything goes. The capability you are paying a premium for today, whatever it is, is on its way to becoming a free tab in someone’s browser. The only durable question is what you choose to do with it once it gets there.