When you ask your coding agent to research a topic, do you read every line of every paper it comes back with?
Before writing, you memorized it or you lost it. Writing made knowledge portable but scarce. Retrieval meant a trip to the monastery. Print made it abundant; "smart" shifted from "can recite" to "has read." Internet search made it findable, but you still had to read the results yourself.
Each transition felt like the end of something. Each one just moved the bottleneck, and the skill that compensated for the old one became a hobby. From recall to access. From access to volume. From volume to synthesis. At every step, language compressed knowledge for the retrieval speed available. Agents made retrieval nearly instant. What happens now?
Retrieval got fast. Not "Google it in a minute" fast, where you still needed the right query, and look through the results. Actually fast: load the corpus, ask the question, get the answer with provenance. The agent holds the proof and you hold the pointer.
You don't need to understand the Bregman divergence generalization of Voronoi diagrams. Just that it exists and that it's relevant when someone questions the Gaussian decay assumption. When the moment comes, you point the agent at the literature and ask. Who cares if you actually know it, if you can answer the question?
Call it the next transition. The bottleneck moved again, from volume to relevance.
The purpose of reading used to be broad: become informed, absorb the field, develop taste. Now it's pointed. You start with a question, and the corpus plays a supporting role.
A colleague emails about embedding-space auctions: "does willingness-to-pay actually drop off with the square distance, or is that just for the math?" The old reader searches their memory, maybe rereads a paper, drafts a reply from recall. The new reader loads the relevant posts and the mechanism design literature into an agent. Sourced answer, seconds.
You don't need to read a paper to extract claims, file them, dedupe, flag disagreements. Reading is compilation of knowledge. Comprehension optional.
Something displaces the content in your head: the index over it. Which ideas connect to which. When a question is worth asking, and which corpus to point at. That sense of "this claim smells wrong" that triggers a deeper look.
Experts always had the map of what exists and where the interesting edges are. Their advantage was always the index.
The data our brains can hold was always tiny. We compensated with lossy compression and hoped the important bits survived, but now the uncompressed version is a query away. Does it matter if you understand the proof, if you can retrieve its provenance just as quickly as the person who memorized it?
Most text is still designed for sequential comprehension. Papers have abstracts, introductions, related work, methodology. Blog posts have arcs. Textbooks have chapters. All of it assumes a reader who starts at the beginning and builds understanding incrementally.
The new reader doesn't do this. They arrive with a question and need the answer. The twenty posts between the question and the answer are infrastructure they'll skip, because reading them would be an inefficient caching strategy for a problem retrieval already solved.
Writing for the new reader means writing for retrieval. Each claim self-contained enough to be extracted. Each dependency explicit enough to be followed. Each post a module, not a chapter. Vibelogging is the write side. New reading is the read side. The same corpus, accessed by inquiry.
The blog that answers a colleague's question faster than they can read it is a queryable system. The writing changes, the access pattern changes, and so do we. Are you a new reader?