One question sits under everything here: can discovery itself be engineered? Not the lucky break, not the lone genius. The process that turns a question into a finding, made reliable enough to repeat and to teach. Take that seriously and two consequences follow. A process a person can run is one a machine might run, which is the AI thread. And once a machine runs it at almost no cost, the only thing still scarce is the judgment to tell a real finding from a confident wrong one, which is the economics thread. Discovery is the core; AI and economics are where it surfaces.
The map that was right last year quietly stops fitting, and a team keeps planning off it until something breaks. We treat ignorance as a thing with shape you can work on, and inquiry as moves you run on purpose: imagine, experiment, describe, communicate, with verification keeping the picture honest as reality moves.
Today's AI answers the question you bring it. The harder thing is an AI that asks its own, runs the inquiry, and does not just game a metric on the way. We build that: judging the process and not only the answer, a reward that resists gaming, and learning-theory foundations a machine can check.
When the machine runs the work for free, the one scarce thing left is the judgment to catch it when it is confidently wrong, and the bill for missing that lands on whoever could not argue back. We name the hidden bill (automation debt), the discipline for paying it down (decision engineering), and the economics of investing against ignorance itself.
The three threads are not parallel by accident. An epistemology that cannot be implemented stays decorative; an AI without a clear epistemology drifts; an economics without either is fiction.