Loading slide
Loading contents...
Language models and agents are changing what work looks like. Not all at once, not everywhere, not for everyone, but really, and the change is still unfolding.
The pattern so far: tasks that used to need specialised skill became easy to approximate with a well-aimed prompt. First drafts, code skeletons, research summaries, translation, reshaping data. For people who know how to work with these tools, all of that got faster.
But notice the careful word, approximate. This is a shift in what one person can do, not a replacement of the person. The model can draft, but cannot decide what is worth drafting or whether the draft is any good. It can surface patterns, but cannot tell you which ones matter. It can help you go faster, but cannot tell you where to go. The judgement stayed with you.
Which work this touches hardest is still being measured, and the early findings are not the ones most people would guess. The International Labour Organization studied how exposed different occupations are, and clerical work came out highest: the desk tasks that felt brainy but were largely routine, transcribing, summarising, first-drafting, basic lookup. Exposure rises with how much of a job is done in words and numbers, and falls with how much of it is done with your hands. Around one in four workers worldwide are in an occupation with some exposure.
Two cautions, because this is where confident predictions tend to outrun the evidence. The same study found exposure increasing in skilled professional and technical roles. Deciding what matters may still rest with you, as the last paragraph said, but that is not the same as your occupation being untouched, and the comfortable version of that story does not survive contact with the data. Second, exposure is not replacement. The ILO's own conclusion is that most jobs will be transformed rather than eliminated, because most jobs are a bundle of tasks and only some of them are exposed.
So the durable skill is not coding, and it is certainly not being able to derive a transformer from scratch. It is fluency: understanding what these systems really are, what they can do, where they fail, and how to tell a solid answer from a confident-sounding wrong one.
Which is worth saying plainly, because it is about you. That fluency is exactly what this course has been building, slab by slab, from the transistor onward. You did not learn to use a particular app that will look dated in a year. You learned how the machine underneath actually works. That is the thing that lasts, and you have earned it.