Agentic AI — S2 · 3/10: Deskilling
A draft regulation is put out for consultation. Its summary will cover the articles, but it may leave out the recitals. Yet the recitals are often where the legislator explains why it is legislating.
When we do our own regulatory monitoring, we look at the whole text, especially while it is still a draft. We try to work out what is most likely to end up in the final version. That means grasping the legislator’s intent and the stakeholder forces at play, which are sometimes opposed. This intent can be read in the recitals, but also in an idea that recurs from one chapter to the next or sits tucked away in an annex.
That kind of monitoring had its limits. It relied on newsletters and alerts published by the authorities, and only the texts deemed important were read. Depending on the size of the company, a dedicated unit handled it, or everyone did it for their own scope. It was up to the reader to decide what deserved to be read in full. And who relishes the thought of reading a dry text of more than a hundred pages, sometimes written in a foreign language?
Today, regulatory intelligence is increasingly automated. Texts are collected, sorted, summarised. What is deemed important is now determined by the tool. What we read is the summary. We have gained time and simplicity.
So we read the texts themselves less. We no longer try to connect the original intent with the compromises that were adopted. We are less exposed to repetitions, nuances, exceptions and weak signals. We end up understanding less.
Once we know the tool will retrieve the texts and summarise what matters, the effort disappears, and with it, progress.
Skills we no longer practise do not disappear overnight. They atrophy through lack of use.
That is how AI stops augmenting expertise.
The more we rely on the tool, the more we unlearn how to decode texts and what they leave unsaid.
What insight have you gained from reading a text in full that no summary could have given you?