Using a different AI model for each task does keep costs down, as long as the handoff between them costs less than the work you moved. Sending forty pages of meeting notes to a small model for a summary and saving the frontier model for the strategy memo that follows is sensible routing, but the memo writer starts with none of what the summarizer read, so you paste the summary back in, explain the project again, and pay frontier-model rates for every token of the catch-up.

The saving holds when both models read from the same context instead of each starting cold, because then the memo writer spends its tokens on the memo rather than on catching up, and the summary only costs you once. Fulcra keeps that context on a timeline you own, so whichever agent takes the next task picks up where the last one left off.

Try it out for free by pasting this to your agent: https://docs.fulcradynamics.com/agent-get-started.txt