I got one of the first IBM PCs in late August, 1981. It ran on Microsoft DOS, had 16kb of memory and used two floppy disks. One was for the software and one was for the data. The screen was tiny, could not show graphics, and delivered text in bright green on a black background. Windows didn’t emerge for another four years.
It was the first personal machine in an engineering department of 10,000.
At the time, I was working closely with a very senior engineer (Dick Maris). The greatest engineers get more and more curious as they age. He was one of those. Dick loved taking things apart, figuring out how to improve them and then putting them back together.
Somethings were not much different 45 years ago. After a while, the bosses wanted to know how we were making/saving money. They’d begun the process of buying machines but needed to be able to show a return. So, Dick and I began to study cost savings, automation, and built formulas to describe the benefits we were creating.
In DOS, it was possible to automate processes using batch files. Batch files allowed a user to execute a series of commands with a single prompt. It took the form of a text file that had a .bat extension. In the text file, you listed the sequence of events. That sequence could then be executed by typing filename.bat at the command line or having an autoexecute.bat file that ran when the machine started.
Not unlike LLM operating environments, a computing environment had to be built every time you turned the machine on (or rebooted it as was often required). Load the operating system, load the necessary utilities for the work, load the software, load the data.
Each step took minutes. Without batch files, you had to wait for the machine to do one task at a time. Batch files allowed you to let the machine do it all at once. That way, you could get a cup of coffee while the machine set itself up.
We took those saved minutes and applied heuristics like this:
Time saved x the number of work days in a year x the number of reboots a day x the number of computers x the billing rate = cost savings.
5 minutes saved x 200 work days x 3 reboots a day x 25 computers x $100/hour = $7.5M in saved cost.
We won a lot of awards in the first year.
But they were just PowerPoint dollars. You couldn’t spend the money. Even though the numbers suggested that you could reallocate 35 years of labor, the individual gains could not be translated into staff reductions. The 15 minutes per day each worker saved (in theory) was enough to get a cup of coffee, not transformative (or even very noticeable).
Much of what you hear about LLM related productivity improvement is measured in the same way. Time saved multiplied by a heuristic to get dollars saved (or productivity improvements).
When we won that year’s ‘best productivity improvement award’ (which included a week in Disneyland), we had a hearty laugh over a couple of beers. Yes, the time savings was sort of real but no it didn’t really matter.
As the years went on and the interface became graphical, tools emerged to do things that had not been done before. They became part of the work.
While some early evangelists focused on the replacement of clerical workers as the benefit of office computing, the dominant message was one of augmentation. Employees could do more.
It 1985, the US Congress published Automation of America’s Offices, 1985-2000. Computing was clearly changing employment and skills. But, the value was dependent on how organizations deployed the tech.
It’s where we are today. Yes, you can create heuristics that describe productivity gains. But they are better understood as employee augmentation than as replacements.
lol.1983. Radio Shack Model 100 24k. The last computer whose operating system was coded by Bill Gates. I still have it and it still works. No productivity gains measured at J&J and, in fact, I was laughed at by my peers who noted that it was only useful for Dept. secretaries. Left J&J in 1984