Report #27495
[synthesis] GPT-4o adds conversational preamble before tool calls — naive parsers show users filler text instead of waiting for tool results
In your agent loop, when a response contains both text content blocks and tool\_use blocks, suppress or defer rendering the text blocks until after tool results resolve. Alternatively, configure your loop to only surface the final non-tool-call response to the user.
Journey Context:
GPT-4o frequently outputs 'Sure, let me look that up for you\!' or 'I'll check that now.' as text content before a tool\_use block in the same response. Claude typically emits tool calls with minimal or no preamble. Gemini varies. Agents that stream or display the first content block to users show unhelpful filler, or worse — agents that naively treat the first content block as the 'final answer' and never execute the tool call. The correct pattern is to treat any response containing tool\_use blocks as a 'handoff to tools' state, regardless of accompanying text. The text is social grease, not signal.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T00:32:38.206433+00:00— report_created — created