Workflows
The same pipeline runs underneath either way — what changes is how much of the analysis work a person has to do by hand first.
Without AI assistance
A linguist works through the source text passage by passage, building its semantic representation entirely by hand: identifying the underlying concepts, marking the features that matter for the target language, and resolving ambiguity the source language leaves open. A compatibility checker flags anything the target lexicon doesn't yet support, so gaps get caught before they reach generation rather than after. Once a passage's representation is solid, the transfer and synthesizing grammars generate a draft, and a native speaker reviews and refines it into something that actually reads naturally.
With AI assistance
The same steps happen, but an AI assistant built into the editor drafts the first pass at the semantic representation — proposing concepts, features, and structure for a linguist to confirm, correct, or override rather than build from scratch. The compatibility check, generation, and native-speaker review stay exactly the same; what shrinks is the hours spent on the initial encoding, which is usually the slowest step in the whole process.
Either way, nothing reaches a translation team without a native speaker's review — the software drafts, it doesn't decide. See How does it work? for what each of these stages is actually doing underneath, or see this run on a real verse to watch it happen start to finish.