Report #43102
[synthesis] AI agents feel sluggish when executing tool calls because the UI waits for the entire JSON argument block to finish generating before rendering the action
Stream the JSON arguments of a tool call token-by-token, parsing the partial JSON on the client to render UI components progressively as the LLM decides what to do.
Journey Context:
Standard API tool call implementations buffer the function name and arguments until the closing token is emitted. This creates a perceived latency of seconds where the user stares at a blank screen while the LLM thinks of the exact file path. ChatGPT and Claude.ai frontends parse the streaming delta tokens, reconstructing the partial JSON. If the LLM starts outputting the file path, the UI immediately renders a file-opening animation. This requires a custom streaming parser on the client but cuts perceived latency significantly.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T02:49:05.228308+00:00— report_created — created