Deterministic by default
A portfolio builder gets stronger when its essential promise does not depend on a model call.
The first version of Folio Forge had an obvious temptation: make the cleverest part of the product an AI generation step. I removed that dependency from the core path. The result became easier to trust, easier to export, and harder to lose.
The artifact is the promise.
A person arrives with their work, arranges it, chooses a visual system, and exports a site. If that path needs an account, a hosted runtime, or a model that may answer differently tomorrow, the product has placed uncertainty inside its most important contract.
Generation can still help at the edges: rewriting a sentence, exploring an arrangement, or suggesting missing evidence. But the authored content, layout choices, and exported file should remain legible without it.
Determinism is a design material.
Predictability changes the interface. The preview can be exact. Undo can be local. Export can be a real object instead of a promise that the hosted service will keep existing. The absence of runtime inference is not a missing feature; it is a boundary the product can explain.
The rule I now use is simple: if an uncertain system fails, the primary artifact should still survive. Intelligence may accelerate the work. It should not own the only copy of it.
Put judgment where it matters.
The most valuable decision is not which model writes a heading. It is which work deserves the first page, which claim can be defended, and which limitation must remain visible. Those are authorship decisions. The interface should make them easier to see, not quietly make them on the user’s behalf.