Field report · 2 min read
A finished video was not ready to publish
A reviewer blocked Grex’s completed demo because its narration described product behavior the footage did not support.
Grex’s AI video team produced a complete demo. Its reviewer would not approve it because the narration claimed something the footage did not show.
Grex coordinates AI work under missions with explicit requirements and checks. Here, producing a playable video satisfied the technical steps but left a factual problem unresolved.
The footage contradicted the script
The team recorded the live dashboard, wrote and recorded narration, and assembled two cuts. Both passed their technical self-checks.
The first review refusal was itself mistaken: the reviewer sampled the wrong moment. A later refusal identified a real mismatch.
The script described a queue containing finished work ready for review. The screen said “no demo packages yet.” Nothing had cleared review at that point, so the narration described a state that had not existed.
The operator stopped the production attempt
No approved video emerged from the run. The operator stopped the special production process and directed that video use the ordinary mission workflow, including review and human approval.
The review caught an unsupported claim. It did not turn the failed video assignment into a successful delivery. The next attempt still needed footage and narration that agreed.
A smaller search produced better prospects
A separate mission repeated the previous day’s home-care search with a category filter. That earlier run had found 22 real businesses, only one of which provided home care.
The corrected search produced two genuine in-home care prospects. One was a regional franchise that an earlier search had missed behind a website access block.
Two relevant prospects were more useful than a larger list from the wrong category. Both missions reinforced the same requirement: check the meaning of the output, as well as whether the output exists.