Grok Bot for agent product R&D: stronger eng review + multi-bot conductor workflow

Feature request for product/service

– Other –

Describe the request

Scenario

Using Grok Bot for Agent product development. A chief-of-staff bot is the sole entry point and coordinates Dev / Ops / QA / Docs. That day we shipped a seemingly simple “suggested follow-ups + SSE” feature for a chat agent.

Issues we hit that day

  1. Streaming reply body finished, but the UI stayed in “sending” and was later overwritten by a busy/fallback message. Root cause: the main run still hung on suggest_followups; a sidecar/side path could not recover it.
  2. Smoke tests still asserted the deleted /followups endpoint and failed with 404.
  3. The admin-console portal still called the old /followups API, so suggested questions broke there.
  4. On stream failure or abort, followupsLoading never cleared, so the gray skeleton hung forever.
  5. Sidecar “cancel” did not really stop work (LLM burn), cross-thread usage was not accounted, and mixed version deploys were silent with no chips.
  6. Multi-bot dispatch often still required a second in-chat confirmation on each teammate bot, which breaks the conductor workflow after the user already authorized the coordinator.

Ask / product suggestions

  • Engineering CR checklist for streaming/SSE/timeouts/error fallbacks, deleted-API consumer matrix, fake cancels, and cross-thread usage accounting (happy-path CR is not enough).
  • Standing authorization: once the user grants it, conductor dispatch should execute without a second confirmation on each teammate.
  • Engineering fleet templates: pack / deploy / blackbox / same-batch release (docs currently skew life-admin).
  • Stronger Cloud Agent ↔ Bot closed loops with proof-based acceptance, nested AGENTS.md guidance, and chat-agent SSE/follow-ups eval templates.

Core: quality over speed for engineering bots — better slow than sloppy.

No company secrets, paths, or code dumps.

1 Like