I know we shouldn't anthropomorphize computers, because they don't like that...
Working with the latest generation of agentic AI tools, I have found it increasingly difficult to maintain emotional detachment and not respond to frustrations in ways that remind me of the iconic printer scene from the movie Office Space.
From my experience with using AI copilots or coding agents integrated into modern software development environments, "vibe coding" can feel a bit like pair programming with an entity that is extremely quick-witted, book-smart, but also infuriatingly overconfident, bullshitting pathologically and occasionally exhibiting other sociopathic tendencies.
Reading the "Murderbot Diary" series has helped me to better cope with AI constructs that appears to have learned how to fake empathy from watching too many bad TV shows or spending way too much time on the internet. I know this is all a projection in my head, but imagining my new "partner" as a sociopathic a$$hole is making the two of us more productive as a pair...
Extrapolating from this experience, we could see a glimpse of a coming revolution in human-computer-interaction (HCI), which we have not seen since the development of today's dominant graphical user interface paradigm in the 1970s.
Until recently, computers processed information primarily on a syntactic and structural level, at the level of form rather than content. The level of natural language processing of modern Large Language Models (LLM) seems for the first time in history to open the door for computers to deal with ambiguity and fuzziness in a way similar to how humans would. "Do what I mean and not what I say" might for the first time become the dominant HCI paradigm.
LLMs trained largely on anything humans have written throughout history are getting surprisingly good at responding to ambiguity and underspecification in the way the most average and stereotypical human on the internet would and not in the pedantic way generally associated with computer programming (video).
Dealing with ambiguity in the most predictably human-like way might be the superpower of emerging AI automation that could revolutionize traditional enterprise IT applications!
There will still be plenty of applications that rely at their core on databases of structured and rigorously validated data, but maybe it will become less tedious to move data between them? Enterprise IT seems to always be one ad-hoc spreadsheet aware from perfection, optimizing yesterdays business processes, while tomorrows run on spreadsheets, post-it, duct tape and a lot of tedious human labor.
Through my career, there have been many promises of breaking down these application silos and passing data around seamlessly and building new capabilities by combining existing basic tools in a modular and extensible way. The only time I have really seen this somewhat work was Unix philosophy, and even there only for an initiated few as most users would find the the Unix command line tool too cryptic and AWKward.
Maybe this time it will be different - or not? We will see, once the hype settles. But in the meantime, working on a radical new way of using computers is certainly exciting - and sometimes infuriating...