Will Larson recently targeted the real constraint that AI workflows need to overcome to reach full potential — overcoming the constraint of judgment. AI workflows can speed up our work (saving time), and they open the door to tasks in parallel (preserving attention). However, there is still the need for judgment to maintain these workflows as reliable.
Larson’s developed the idea of “datapacks”: AI-friendly embeddings (XML or similar) of traditional documentation. Documentation preserves domain-specific knowledge beyond direct interactions via the original author; Datapacks can embed this expertise into AI to scale judgement.
There is another consideration to expand upon this. It’s one thing to transform the original knowledge (e.g., documentation) into AI-friendly format (e.g., XML), and it’s another thing to ask: What’s the best way to preserve the original knowledge itself, regardless of the end format?
It’s undervalued to look at ancient sources to inform these ideas. This requires admitting that sometimes our advances in tooling isn’t necessarily an advancement in practical wisdom.
One overlooked idea, that I think applies directly, is the scholastic method.
The scholastic method emerged in the 12th century as a tool for synthesizing and systematizing civil law. Its most famous application came with Gratian’s Decretum (c. 1140), a monumental attempt to reconcile centuries of contradictory Church canons. For each apparent contradiction, Gratian applied a consistent format: here are the two authoritative sources in tension — here is the distinction that resolves the contradiction — here is the principle that now applies going forward. It didn’t just record conclusions. It recorded the reasoning behind the conclusions, including what was considered and why it was set aside. In short, it was a technological breakthrough for preserving the original intent behind a ruling so that future readers could most clearly understand what was meant.1
Perhaps this ancient technology can be coupled with the bleeding-edge. The translation from documentation to an XML datapack is promising, but we could also improve what is being formatted as AI-friendly: the documentation itself. An updated version of the scholastic method could add a repeatable structure for judgment-maximizing documentation.
OpenSpec gets close to this format with their WHEN/THEN prompt templates. However, they’re missing the WHY and the clear statements of what it is NOT intended to do, which the scholastic method preserved.
Maybe this extension — of writing and not immediately tooling — will be the highest leverage move to progress against the constraint of judgment.
Grant, God and Reason in the Middle Ages, 65-66, 76-80.