hi community,
been battletesting this new AI-driven database approach on ~5–6k records, adding ~1k daily. hitting some sharp edges with field agents.
1. “run automatically” is too aggressive
it keeps re-running. great for freshness, bad for cost.
with real volumes this becomes monster token drain.
what’s missing:
-
run automatically once (if empty)
-
then freeze unless explicitly re-triggered
more like a computed default than a live formula.
2. no clean automation trigger
still can’t call a field agent directly from automations.
so no way for controlled execution
ideal model
-
auto-run once → initialize
-
automation trigger → intentional re-runs
first fixes cost
second fixes control
together makes field agents viable at scale.
