Most A/B testing is organized amnesia.
You test two subject lines. One wins. Six months later you spin up new variants, and run the exact same test because nobody remembered.
Fix: store every variant and result in a database. When the AI generates new tests, it reads that history first.
So instead of random variants, you get variants that build on what you know.
'You don't have to remember what A/B test you did six months ago. You have it all stored.'
Testing that compounds instead of resets. Big difference.