Building Enterprise Agents: How to Implement a True Pause for Human-in-the-Loop
Reasonable-Lack-7701 · reddit · 2026-08-04
The author shares lessons from spending two years building a human-in-the-loop (HITL) approval system for an enterprise agent platform. The core insight is that a genuine approval step must act as a hard block in execution, not just an after-the-fact log entry.
The workflow: an agent proposes a tool call -> execution suspends -> a human reviews the exact parameters inline (which are editable, not restricted to simple approve/reject) -> the call fires upon confirmation. The author stresses that rejection or editing must be a first-class outcome the agent can react to, rather than a bolted-on exception path.
Noting that there is no settled industry pattern for cleanly suspending and resuming an agent mid-tool-call, the author asks the community for architectural approaches: full request/response blocking, a queue polled by the agent, or an event-driven model?
More from coding & agent
- Google Paper: Financial AI Agents Excel at the Past, Fail at the Future — rohanpaul_ai · 2026-08-04
- Developer Shares Summary Diagram of AI Agent Harness Evolution — bingxu_ · 2026-08-04
- ChatGPT is Breaking Codebases: Devs Complain About Model Regression — wowa93 · 2026-08-04
- Quickly Turn Any Website into an API or MCP Server Using DevTools — dsp_ · 2026-08-04
- Anthropic Engineers Demo Multi-Agent Loops: Full App Built from Scratch in 40 Mins — mathemagie · 2026-08-04
- Running Agents in Prod for a Year: Auditability Trumps Model Capability — KimLikeJ · 2026-08-04