Definition
Plain language
The usual way of chaining AI agents, where each one finishes its whole answer before handing it to the next.
As stated in the literature
The standard sequential multi-agent protocol in which an upstream agent emits its complete reasoning before a downstream agent begins; contrasted with streaming, where partial reasoning is passed as it is produced.
Why it matters: It is simple but slow, since every agent in the chain sits idle until the one before it completely finishes, motivating faster streaming alternatives.
For example, a research agent writes its entire analysis, and only once it is fully finished does a second agent start drafting the summary.
Heard on the show
“The authors call this "generate-then-transfer.”Episode 116 — Why Streaming Half a Reasoning Chain Beats Sending the Whole Thing