As Chris Aldridge says, for centuries the Zettelkasten approach was the standard and universal method for producing books and articles - until personal computers took over. Nearly every serious work ever published before the 1980s was drafted either with index cards or paper slips, or else with notebooks in a commonplace style. Every writer had their own take on these two options, but that’s what they all used. Then, in a single decade, word processing software took over. These days, most writers use something like Microsoft Word or Google Docs (just try persuading your publisher you’re not giving them a docx file). Scrivener became popular because it critiqued the ‘endless roll of paper’ model and reverted to an index card interface of sorts. But it remained a niche.Today, you either thrive on that word processor model or you don’t. I really don’t, which is why I’ve invested effort, as you have, in researching previous writing workflows, older than the all-conquering PC of the late 1980s and early 90s. At the same time, new writing tools are challenging the established Microsoft way, but in doing so are drawing attention to the fact that each app locks the user into a particular set of assumptions about the drafting and publishing process.The current academic scene is a brutal war to publish or perish. It’s not unusual for a researcher to write or co-write 30-40 peer-reviewed articles per year. General publishing is also frenetic. In the UK, 20 books are published every hour of the day. It all makes Luhmann’s ‘prolific’ output look lazy. Now though, AI is blowing the entire field apart. From now on, prolific writing is what computers do best. There’s no reason not to publish 20,000 books per hour. Soon enough, that will be the output per ‘author’. Where the pieces will eventually land is anyone’s guess. For example, the workflow of the near future might involve one part writing and nineteen parts marketing. Except that AI has got that sewn up too. Meanwhile, until the world ends, I’m just having fun doing my thing.

For example, within the literature on 44data centres, top-down models that project energy use on the basisof increasing demand for internet 45servicestend to predict rapid global energy use growth, (Andrae and Edler 2015; Belkhir and Elmeligi 462018; Liu et al. 2020a),whereas bottom-up models that consider data center technology stocksand their 47energy efficiency trends tend to predict slower but still positive growth (Hintemann and Hinterholzer 482019; Masanet et al. 2020; Shehabi et al. 2018; Malmodin 2020)