The Scholastic Trap

You constrain the agent via the harness, build evals you do not fully trust and ship anyway. The Stoics called this acting under uncertainty. It is not new, but it is where technology and engineering live now. This same spirit, or phenomenology, is part of the shift from Aristotle to Stoicism: Aristotle still writes from a world oriented toward ordered flourishing; a few decades after his death the Stoics inherited a more unstable Hellenistic world and turned philosophy toward disciplined action when outcomes are no longer fully in your control. In that sense, the current phenomenon in "agentic engineering" is more Stoic than Aristotelian: you act with structure, constraint, and judgment, not with guarantees. Aristotle's own works did not survive the journey. The Latin West lost nearly everything except his logic. When the rest returned centuries later, the Church controlled the re-entry.

Throughout history the power struggle has always involved technology and knowledge. The power systems took the model framework the way the medieval Church took Aristotle. Like Aristotle, the harness ("Claude/Claude-code") was deemed a "national threat" just as the University of Paris condemned him in 1210. By 1255, Aristotle had become required reading in the Paris arts curriculum. That turn took roughly forty-five years. Our cycle is faster. Much faster.

The Church had initially confined Aristotle to Latin. Today, transformers, chain-of-thought, tool use, RLHF, all things published in the open but requiring massive amounts of resources to make usable, are shipped behind an API key. The genealogy is invisible at the receiving end. You are not using the original framework, just as you are not reading Aristotle from his own words but rather in Latin, which had been translated from Arabic from Greek. Frontier models are hence like a translation of Aristotle; closed to anyone who is not affluent enough to, in the case of the scholastic era, speak Latin. We are using an institution's translation of something, and the institution has decided that its translation is the only legitimate one.

A "Claude-bro" may ask, "Are you using the right frontier model?". This is not a technical question. It is a (vernacular) "Latin question." The work done outside the institutional language thus registers as vernacular. But the vernacular is not lesser. It is different. Local models fine-tuned on proprietary data, running air-gapped, inspectable. Not necessarily inferior. They are work the institutional medium cannot afford to do: someone doing the work without needing the institution to mediate it. That is what the Scholastic framework could never tolerate: independence.

Frontier models' economy will not collapse because it is wrong. The frontier framework produces genuine results. It collapses because the institution could no longer control the means of verification. Luther and the printing press put the text in anyone's hands. Galileo's telescope let anyone check the claim. The cracks are already showing; the Scholastics began publishing in the vernacular: Google published Gemma 4 under an Apache 2.0 license last week.