We spent months talking about artificial intelligence, In many cases, bar conversations are discussed. We speak with great light of his promises, of total transformations and narratives that mix certain predictions with a certain taste for dithyrambs. And with greater freedom, we allow ourselves to diagnose disasters of all kinds, from our impact on cybersecurity to the workforce.
As for me, I bide my time by proposing to speak of a more balanced form, and to grace what I like to call account of promise. That is, put numbers on the distance you have between what is announced and what is actually used, or on direct and direct impact. And this appears when you exercise, e.g a broad stream of artificial intelligence, acting in silencein a real fountain of works.
Investigation publicly available we argue its implications for the workforce. The study analyzes actual usage, starting with actual conversations with the language model during December 2024 and early 2025, mapped to the O*NET taxonomy. Of the 19,500 tares that this system takes, only 3,514 show non-trivial use. Hasta ahí, everything can seem reasonable. The coup combines distribution with targeting: the market median weight is 0.006% and the mean is 0.028%. I say it differently: most observed works live in a large area, with anecdotal cases a minority concentrates actual adoption.
El paper It will be clearer when you segment in more detail. Only 43 areas, which is more than 1.2% of the total number of areas analyzed, have high utilization. At the other end, 3,353 tare, or 95.4%, is down 0.1%. And there is a fact that is worth repeating in detail, because someone reviewing an invoice line by line: the most common activity, modifying existing software to fix bugs or improve performance, accumulates 4.79% of all interactions. One charge is 5% of the total consumption. That’s the broad thread of AI in its purest form, without the baroque-technological rhetoric obscuring it.
When one looks at which operations form the basis of this distribution, the owner becomes absurdly monotonous. Code cleanup, bug fixing, troubleshootingwriting and editing programs, technical maintenance… Anything that is well expressed in text can be done quickly and fits seamlessly into a conversational interface. When one enters the broad chain, there appear to be areas with weights on the order of 0.00195%: price negotiation, interpersonal coordination, interdepartmental collaboration, to areas of high clinical responsibility. It’s not that AI doesn’t exist, unfortunately; The point is that for conversational use, which is the name, it is marginal. This is also part of the promise bill.
A jump in employment levels defies the same logic. From 36%, Alrededor keeps AI present on less than 25% of its dimensions. Data sounds great until it is understood with the following: less than 4% achieve high integration, up to 75% of areas with observable use. Let me put it another way: adoption is lengthening, but not deepening. The head grows; cola is spreading. In the middle is a huge gray area that supports day-to-day work as soon as it is touched.
Including the form of acceptance AI reacts to this geometry. 57% of interactions are classified as como increase and 43% as automatic. I prefer direct replacement; another internal reconfiguration that breaks like this. Artificial intelligence enters where it can speed up, improve and redistribute very specific activities, many of which are the least glamorous and most repetitive, which is rarely the case when they are the protagonists of foreign discourses but support daily operations.
Yes, take it away paper It is an antidote to discursive inflation. So we have more of the broad scope of AI and accounting for promises and less arguments influencers and visionaries of dudoso origin.
Leave a Reply