The Wall Street Journal has breathlessly headlined that Prompt Engineer (“The Hottest AI Job of 2023”) is “Already Obsolete.” No-one should be surprised – but looking at the last seven decades of people asking, ‘What’s the exotic skill this year?’, may offer guidance on how to avoid becoming road kill on that highway.
Today, writing a well-structured and thoroughly scaffolded prompt might still inspire a label like Prompt Engineer, but ‘spreadsheet jockey’ also used to be a term implying valuable expertise: according to Google Ngrams, occurrence of the latter phrase in the corpus of Google Books has declined by about one-third since 2010. The even-shorter life of Prompt Engineer began, according to Google Trends, with its appearance from nowhere and an abrupt but brief spike in April 2023 – then another spike in October 2024, but a decline of more than 60% in search traffic for that phrase since then.
Blast from the past – it may be difficult to believe that when spreadsheets were a new thing, knowing how to use them for anything more than simple tables of summed rows and columns was a distinction. “Are you a ‘wizard’?” was the actual question that I was asked by an economics Professor, sometime around 1984 or ’85, when I embedded a spreadsheet-generated bar chart in a document on what was then a high-end Macintosh 512K. How quaint.
Quick decline
The decline of the hype around Prompt Engineer is happening more quickly than I might have predicted a year ago – but it’s just the latest chapter of a saga of rising levels of abstraction, and concurrent de-valuing and de-glamorizing of briefly exotic skills, that we’ve seen taking place since the 1950s. We’re seeing the next, predictable, step from machine-level expertise: people used to optimize instruction sequence to match the rotational speed of the magnetic drum memory. Seriously.
In drum memories, information was only available at one point in rotation, therefore the computer might have to wait for an entire revolution to retrieve a piece of data or instruction. Users often sought to optimize these retrievals by distributing instructions and data so that minimal time was lost to waiting for the drum to rotate. Such optimization was completely hardware specific…
Moving on to virtual-machine understanding: the C family of programming languages (C, C++, C#, Java etc.) show their heritage of being designed to make use of the PDP-11 and later the VAX processor architectures, even though Java exclusively runs on the “Java Virtual Machine” – which has to be implemented, usually in some other language like C, for whatever hardware is actually being used.
As suggested in the IBM developerWorks journal:
Understanding bytecode and what bytecode is likely to be generated by a Java compiler helps the Java programmer in the same way that knowledge of assembly helps the C or C++ programmer.
Then further progressing toward totally abstract representation of code as a form of data, and data as an entirely machine-independent structure such as the linked lists of Lisp or the terms, lists and trees of Prolog. As Edsger W. Dijkstra in his 1972 Turing Award lecture:
LISP has jokingly been described as “the most intelligent way to misuse a computer”. I think that description a great compliment because it transmits the full flavour of liberation: it has assisted a number of our most gifted fellow humans in thinking previously impossible thoughts.
And finally to the point of today’s non-procedural ways of telling a machine what you want, and letting invisible mechanisms figure out how to do it: things like SQL for databases, or less well-known ‘production system’ technologies like OPS5.
OPS5 uses a forward chaining inference engine; programs execute by scanning “working memory elements” (which are vaguely object-like, with classes and attributes) looking for matches with the rules in ‘production memory’. Rules have actions that may modify or remove the matched element, create new ones, perform side effects such as output, and so forth. Execution continues until no more matches can be found.
Perhaps I should assign the ‘finally’ point to spreadsheets, in which much of the world’s computing gets done by people who totally do not think of themselves as “programmers” or “coders” or even “business analysts.” Spreadsheets are a means of expressing what we know, and asking questions about what that knowledge implies. It’s not wizardry any more – and in a globally connected world of magic black glass portals to fact and function, it’s been a long time since people were impressed by mere complexity of ‘how?’ without utility of ‘why?’.
No one, not even me, optimizes the layout of a spreadsheet to minimize recalculation time in 2025 – although I did invent a spreadsheet benchmark, the Savage Spiral, designed as a torture test for both speed and correctness back in my 1990s period of reviewing software (when you could easily measure recalc times with an ordinary stopwatch). Today’s offerings come so close to instantaneous accuracy that it seems pointless to measure the narrowing gaps, or for almost anyone to bother knowing how to make those gaps still smaller.
The right questions
The current Salesforce Trailhead offerings for prompt learning may represent the dot on the time line, after which we’ll someday say that:
Understanding the inner life of a prompt was actually important back then – but today the Reasoning Engine figures out what’s relevant, and asks the right questions to get there faster.
Like optimizing compilers, like inference engines, like spreadsheet dependency trees today.
There’s a slide that I first used in 2016, when people talked about “coding camps” and suchlike as if they represented the golden ticket to a high-wage career. I observed at the time that people spoke of “programming” as if a driver’s license were an entry point to a career as a Formula 1 driver (making about $30 million a year), when it is much more likely to lead to something like a long-haul truck driving job ($90k/year) or even an Uber driver (about $30k/year). Similarly as to being a coder – with Prompt Engineer quite plausibly headed for the same off-ramp.
How does an aspiring professional still working on a degree, or a mid-career contributor who’s not eager to retire, avoid going under the wheels of this still-accelerating juggernaut? The tempting error will always be the mastery of the currently novel (and therefore scarce) skill. A strategy for success may be to ask, not how that skill is being used, but what previously unsolved (or even thought to be unsolvable) problems are most often being mentioned as the domains that will be revolutionized by its application.
Not mere ‘computerization’ or ‘programming’ or even ‘Prompt Engineering’ – but (for example) ‘climate modeling’ or ‘cancer therapy’ or ‘personalized teaching assistance’. None of those looks to become an obsolete interest, any time soon – and these are just a few of the goals that will always engage new skills, again and again and again.
