AI recommendations are not approvals: 3 controls in FinSentry payment-agent demo
Proper-Spread-35 · reddit · 2026-10-12
A developer shares FinSentry AI, a side-project demo (synthetic sandbox data) arguing that an agent's explanation must never silently become authority to act. Stack: React/TypeScript, FastAPI, LangGraph, PostgreSQL; domain services own case state and approvals while the agent runtime only coordinates advisory steps.
Three concrete controls:
- Abstain on missing evidence — a plausible model explanation is not clearance.
- A request is not an approval — maker/checker with self-approval restrictions; a "human in the loop" label alone doesn't separate requester from approver.
- Approve an exact action, not a vague outcome — the demo approves maintaining a hold, releases nothing, and the audit trail shows exactly what changed.
Proposed production acceptance tests cover missing inputs, same-user approvals, evidence changing post-request, and duplicate execution; next boundary to explore is stale approvals.
More from coding & agent
- David Fowler: build CLIs for agents to explore systems and fix bugs autonomously — mariorod1 · 2026-10-12
- rtoml v0.14 Adds Python 3.15 Support, i128 Ints, and Big Speedups — samuelcolvin · 2026-10-12
- Reverse-engineering ChatGPT's Intelligent UI: DIL language, server-side compilation and sandboxed runtime — rohanpaul_ai · 2026-10-12
- Reverse engineering leaks three unannounced Grok Bot features, versioned memory leading — bfrench · 2026-10-12
- badlogic demos durable codemode: persisting agent state so long tasks survive aborts — max_paperclips · 2026-10-12
- supermemory packages agent context into one API with memory, retrieval built in — julianweisser · 2026-10-12