One AI Agent vs. Several: What's Different

The newest thing on the checking side is one AI that reads every language in one sitting.

The rest of that side is split by language, each checker kept to its own lane. What none of them can do — it has come up here before — is see a thing that sits across two drafts at once. The one reading the whole set flagged what no split checker had: among the flags, two drafts that had arrived at the same sentence. Inside either language there was nothing to catch; the finding sat between them. Sorting those flags is handled by hand.

Running several also costs something on this end. Fixing a line in one language can turn up a fresh overlap between it and a second one. After every fix, the whole set gets compared again.

The writing is a different story. A slot for a separate writer per language exists and has stayed empty; every language's draft comes out of the same one. The plan pointed at more writers. The running kept pointing back at one.

A single entry on its way out passes through both shapes: drafting alone, checking in numbers, the cross-read alone again.

One-off questions still go to a single AI and stop at its answer.

Whole stretches go by with no one looking in. An AI writes the drafts. No draft has gone up the way it arrived — each takes a final pass first. And handing a step to one AI, or to a row of them, stays a human call.