How many AI experiments should a team run at once?
A team can only run as many AI experiments as it has capacity to finish. Keep a short list with an owner on each one, and compare it against the time and specialist support you really have.
Continuous innovation doesn't mean starting every available experiment. Each one needs enough capacity to evaluate results, support participants and decide what happens next. For each active experiment, list the owner, the unresolved question, the next checkpoint and the work needed to finish. An idea without a responsible person isn't an active experiment yet.
If something can't get a meaningful review, I'd pause or defer it. And don't borrow the same employee's time across several plans as though it were unlimited.
Decide how new ideas get onto the list. A promising proposal may replace a weaker trial, as long as the choice is explicit, and deferred ideas stay clearly separate from commitments. Judge the list by the decisions made. The useful rhythm is investigate, decide, release capacity. A smaller set of completed learning cycles may serve you better than a big pile of prototypes that stay unfinished.