AI as a weapon of mass cognitive destruction

I use AI every day; it’s unavoidable when you create agentic tooling. But something has been grating on me for months, and it isn’t about development: non-technical people are using it to generate far too much slop. Not code slop, but business slop.

Five paragraph meeting agendas. Four-page responses to a planning inquiry, because that way I get “all the facts”. A one-line question turns up dressed as a formal memo with a summary, three bullets of tangentially related sub-questions and a partridge in a pear tree. Documents that used to run a page now run six, padded out with restated context that nobody wrote by hand and nobody reads. Words are now effortless, so people produce more of them–whether or not they still mean anything.

The immediacy is a bigger dopamine hit than you’d expect. Type a prompt, get six paragraphs and two tables in seconds, and it feels efficient. The sender reckons they’ve saved twenty minutes, and from where they sit, they have: what took twenty minutes now takes two.

Except the time didn’t vanish, it shifted and multiplied. Every recipient now has to wade through the padding, work out which sentence actually matters, and mentally rebuild the one-liner that should have been sent in the first place. Sender spends two minutes; ten people downstream lose fifteen each–if they read the whole thing at all.

People in tech love AI because, well, let’s face it, few of them can write; the average coder isn’t very communicative. And if you do spend a lot of time communicating, the illusion above soon has you in thrall.

AI in the hands of people who can’t use it effectively simply dumps cognitive load on everyone else. I’m something of a speed reader (part of my slightly off neurological makeup), and it’s driving me nuts that a context switch which used to be instant now means ploughing through pages of vaguely dressed-up pseudo-facts.

Hilariously, one of the last times I replied to a verbose e-mail – in my usual terse one-liners and bullets – I pointed out that the data fed to the AI that helped create that e-mail was slightly off. The response was baffling: people actually mistook my British-leaning vocabulary… for an AI-generated response. Because, yes, I use em dashes.

The worst part is that management usually can’t tell the difference, and increasingly doesn’t try. More emails, longer documents, quicker turnaround–it all looks like productivity, and output is easy to count in a way quality never is. So the person firing off ten inflated reports looks busier than the one sending a single tight paragraph that actually settles the question.

This runs top to bottom of the org chart, and it comes down to a basic confusion about what AI is for. The point is to save effort–same result for less, or a better result for the same. Instead people use it to inflate the same result into something that looks bigger. Leadership watches the volume tick up and reads it as a gain. All that’s really happened is the work got louder.

And given the constant pressure to “sell”, to be noticed, to “achieve more”, we’re actually rewarding volume and calling it work because the worst KPIs are the easiest to measure: count of emails, length of documents, speed of reply. Just like measuring the number of PRs landed, or lines of code.

Nobody is measuring the cognitive load being dumped on the other end, or whether the message actually landed, or how many recipient-hours were wasted. So on paper the overload doesn’t exist–it just compounds, off the books, while some management dashboard stays in the green–“line go up”, right?

Blaise Pascal once quipped “I would have written a shorter letter, but I did not have the time.” Being both factual and concise is the expensive option, not the lazy one. Halving a document means understanding it well enough to know what can be trimmed away. Getting a paragraph down to the pair of sentences that are really key takes judgement, a couple of revisions and, yes, time. And (this is what really annoys me) we still haven’t nailed the art of the concise slide presentation–instead, we’ve now weaponized it to… a nuclear degree.

And since I’ve mentioned slide decks: taste is one of AI’s main casualties, visual and textual alike. I abhor the sleek corporate jargon now available as e-mail ammunition to everyone–but that deserves its own post. And, like Pascal, I’m starting to feel I didn’t spend enough time trimming this one down…