Three-Tier Architecture for Real-World Agent Action Approval
AiGentsy · reddit · 2026-07-17
When AI agents need to perform real-world actions like sending emails, calling APIs, or transferring funds, simple prompt-based confirmations or Slack notifications are often bypassed by the model or lead to blind human approval. The author proposes splitting the approval mechanism into three distinct layers:
- Policy Layer: Before notifying a human, deterministic rules should dictate whether manual intervention is required (e.g., auto-approving micro-transactions while blocking high-risk ones), thereby mitigating approval fatigue.
- Execution Layer: The approval gating must be structural; operations cannot reach the actual function without permission, and it must "fail closed."
- Evidence Layer: Approval records should contain offline-verifiable cryptographic signatures, rather than self-attested logs stored in an internal database.
Based on this paradigm, the author developed the open-source tool aigentsy, which wraps tools in a single line of code and returns verifiable proofs.
Related event: Approval and Audit Debates for Production AI Agents(6 posts)→
More from coding & agent
- Vovy launches a cursor-following voice agent for Mac, free tier gives 100 credits/mo — Scobleizer · 2026-10-27
- Redditor seeks real-world stacks for a persistent-memory personal AI that answers texts and emails — BrooksomidusGG · 2026-10-27
- Practical tip: use max effort for planning, xhigh for implementation in Claude Code — majidmanzarpour · 2026-10-27
- Chrome team explains WebMCP: agents get tools from your website, not a replacement for MCP — rseroter · 2026-10-27
- Agent-Native apps: why agents bypassing your UI can break business rules — rseroter · 2026-10-27
- 'My intelligence stack': developer splits AI coding work across Opus 5.5, Fable and Astra with human review gates — sh_reya · 2026-10-27