Every time a chatbot gives a wrong answer with total confidence, someone writes a piece declaring AI a bubble about to pop. Every time it gets something impressively right, someone else declares the singularity imminent. Both reactions are getting in the way of the boring, useful conversation we actually need to have.
Public conversation about AI capability tends to swing between two equally unhelpful poles: either the technology is fundamentally a parlor trick destined to disappoint everyone who invested in it, or it’s on a straight, inevitable line toward something close to general intelligence within a few years. Neither framing leaves much room for the messier truth, which is that these systems are genuinely useful for a specific, growing range of tasks and genuinely unreliable for others, often in ways that aren’t obvious until you’ve been burned by them once.
What frustrates me about most coverage of this topic is how rarely it grapples with that specificity. A story either treats a single impressive demo as proof of imminent transformation, or treats a single embarrassing failure as proof the whole enterprise is a mirage. Both are cherry-picking a data point to fit a narrative that was decided in advance.
The useful question is narrower than the exciting one
The genuinely useful question isn’t whether AI will change everything or nothing. It’s the much narrower, less dramatic question of which specific tasks, in which specific industries, are reliable enough today to actually delegate, and which aren’t — and that answer changes constantly, task by task, as the technology improves unevenly across different capabilities rather than uniformly across the board.
I’d rather read ten careful pieces about which specific white-collar tasks are and aren’t reliably automatable this year than one more piece confidently predicting what the entire economy will look like in 2035. The narrower question is answerable with actual evidence. The sweeping one mostly just tells you which priors the writer walked in with.
