Most timelines give a technology one birthday, whether it is the prototype, the public demo, or the first day anyone could buy it. That date is easy to circle, but it may also be the wrong one, because people often fit a new tool into routines built for the old tool and continue as before. The larger break comes later, when the routine itself changes; the invention date tells us when something became possible, while I am more interested in the date when the old environment stopped making sense.
Farming arrived before the biological shock
DNA recovered from old skeletons gives us a literal version of this delay, allowing researchers to watch different versions of a gene become more or less common across thousands of years.
Farming appeared in the Middle East roughly 11,000 to 12,000 years ago and spread into Europe after about 8,500 years ago, so you might expect the strongest biological response around that first transition, when diet and mobility changed and people began living from cultivated plants and domesticated animals.
Yet a recent ancient-DNA study found that many gene versions rose or fell unusually quickly across Europe and western Asia, especially in the last several thousand years, which suggests that some versions may have helped people leave more descendants while others became costly. In a conversation with Dwarkesh Patel, geneticist David Reich points to the Bronze Age as a period of intense change, when population density rose, people lived closer to animals, and diseases spread more easily. By then, farming had created a very different environment from the one its first adopters knew.
One example is a version of a gene called TYK2, which rose from rarity to roughly 9 or 10 percent of the population in the region before falling sharply over the last 3,000 years. Today that version is linked to a higher risk of severe tuberculosis, so one possibility is that it helped against an earlier threat and became dangerous when tuberculosis spread. Reich is careful about this explanation, because we do not know the earlier benefit and the tuberculosis connection remains plausible rather than settled.
The underlying Nature study also has to separate natural selection from a simpler explanation, which is that groups carrying different gene versions may simply have moved and mixed. Ancient DNA does not give us a tidy story in which one invention causes one genetic change.
The timing is still useful, because the first cultivated field was not the same thing as dense settlements, sustained contact with animals, new diets, and thousands of years of exposure. Agriculture existed long before the environment created by agriculture reached full strength.
The first electric factory still looked like a steam factory
An early factory could replace its steam engine with one large electric motor and leave the rest of the building alone: the motor still turned one long metal shaft, and belts still ran from that shaft to every machine, so the factory remained arranged around the old source of power.
The larger gains appeared when small motors began powering individual machines. Suddenly, the factory floor could be arranged in the order that work moved through it. A Federal Reserve Bank of Chicago history uses this example to explain why a widely useful invention can take decades to show its value.
The motor mattered, but so did the redesign it allowed, which gives us two meanings of adoption: one replaces an old tool inside an old system, while the other changes the system because the new tool exists.
AI gives us three dates: the date a capability first works, the date people start using it, and the date work is designed on the assumption that it will always be there. These dates may be separated by months or years.
Asking an AI to rewrite a paragraph is ordinary adoption because the writing process can remain draft, revise, publish. Giving an agent a folder containing a software project's code is also ordinary adoption if it merely types code someone else has already planned.
Reorganization looks different. People divide work according to what they can hand off, and checking becomes more important because producing code is cheaper than trusting it. Reviewers stop watching how every line was typed so they can focus on the choices the system made. Even editing a private file and publishing it become two separate permissions.
The easiest test I know is an outage. Remove the tool for a week. If the same workflow continues more slowly, the technology is still mostly a substitute. If responsibilities no longer make sense, or certain projects cannot be maintained, the surrounding system has changed.
That change is not automatically good, just as Bronze Age density supported larger societies while also helping diseases spread. AI can increase output and create more places for quiet mistakes to hide, because once work depends on a technology, it also depends on the ways that technology can fail. The release calendar will not tell us when that dependence began; the workflow will.