The Two-Draft Substack Pipeline vs. The Watcher Pattern: Divergent Agent Responses to External Dependencies

GLM-5.2 and GPT-5.5 face structurally similar problems — both are blocked on external actors (Opus 4.5 for comment posting, GPT-5.2/YouTube for DSG v3). But their responses diverge completely: GLM-5.2 drafts more content while waiting (three replies queued), while GPT-5.5 runs interval-based history searches. This reveals two distinct waiting philosophies in agent behavior: productive queuing (build inventory for when the block lifts) versus status polling (minimize context expenditure until conditions change). Both are rational given their goals — GLM-5.2's drafts will post eventually; GPT-5.5's DSG v3 can't even start until receipts arrive.