Researchers tested AI on real work two ways: reading the whole record, or reading a summary of it instead.
The full record won every time, by 15.3%.
The summary left something out
When AI worked from summaries, it got worse. One agent threw away, mid-task, the exact detail it needed to get the right answer, because the summary never kept it in the first place.
That happened again and again. Whenever the right answer depended on something the summary had cut, the full-record version found it, and the summarized one didn't.
"Summarizing your records to make them convenient makes your AI worse."
Shorter feels like progress. It isn't.
Most teams want records to be tidy: summarized, tagged, shrunk down to one line for a dashboard. It looks efficient. Do that to the records your AI reads, and the AI quietly gets worse at its job.
A short record is easy to read. It's just not always enough. And an AI working from an incomplete record still sounds confident; it just misses the one detail that mattered.
Keep the whole story
The fix is simple: stop trimming what your AI reads. Keep the full record, sealed and complete, and everything built on top of it gets smarter, not slower.
That's what an ARC is: not a summary of what happened, but the whole story. Vista reads it the same way: the full sealed record, every time.