Everyone in your company has an AI model. Productivity isn’t going up, but there’s a load of documents flying around that imply it should be (many of those written by AI). And for each of those documents there’s a load-bearing triplicate of reasons on how those documents could be improved.
You’ve been AI-DDOSED.
There’s plenty of reasons why people feel the need to spam gibberish AI. Speaking to friends and family shows me this is happening everywhere.
For some, the audience matters more than the work (see social media). Some of these are about looking good and some about not looking responsible if things break.
People like to karma-farm. An AI can turn any gibberish into something important and profound sounding.1
People want to help (and seem to forget you’ve got a subscription too).
People want plausible deniability. “The AI suggested this approach” puts the blame elsewhere.
People confuse output with progress. Generating a robust strategy sure beats doing any work.
For others there’s genuine confusion about the tool and see it as an all-seeing genius and then trusting it to do things it can’t and taking its agreeable output as evidence.
People mistake fluency for thinking. AI is an expert in everything you are not.
People want validation. The model will tell you it’s a good draft. People will not.
People genuinely believe it helped. The first 80% takes seconds, the remaining 20% takes forever.
There’s also the additive fallacy (or more bluntly, “more is more”).
People reward effort by length. A multi-page detailed analysis sounds better than “no”.
People are anchored to “comprehensive.” The AI offers seven considerations and guess what? You can always ask it for more.
People focus on getting their work done, without considering the impact on others.
People externalise the cost. Producing the document took five minutes. Reading it costs the company forty.
People can’t tell what’s signal (or don’t want to make the effort). AI might generate something with some insight, but it’s buried in noise.
In some places, AI is an expected default. It’s not just about keeping pace with your peers, it’s about embracing a fundamental shift in how we approach communication, collaboration, and the very nature of professional discourse itself.
People are paying for it. The company bought five seats per person and three pilot programmes that never ended. Something must visibly be happening, or the budget vanishes at the next review.
People imitate their peers. The VP started sending AI-generated weekly updates. Now everyone does.
Framing this as a systems problem is easy. It’s a textbook externality, identical to the email spam economy. Sending costs nothing, receiving costs the world. Every defence we have (sender reputation, rate limits, filters) exists to claw back some of that asymmetry.
The fix is always the same. Internalise the externality. Make producers pay the reading cost, cap inventory on shared channels, make consumption visible (etc). All require discipline, which means they’ll fail. Try an incentive system instead and Goodhart eats you alive: measure seats, prompts run, or documents produced, and you’ll get more seats, prompts and documents.
Nope.
Remember when digital camcorders were all the rage, and you had to suffer watching someone else’s badly recorded holiday movies? YouTube and Instagram came along and built filtering and reputation on top. We need that for office documents, at least until the likes themselves become the goal and the platform falls apart.
In the meantime, the best defence is asking people about the work “they’ve” produced.
It needs to be socially unacceptable to ship slop2.