Should you let AI agents pay for you? Keep the final approval, argues dev
Efistoffeles · reddit · 2026-11-09
Can you trust an AI agent to plan a family trip and then hand over your card? The author argues yes—but only under one condition: the final purchase decision must stay with you.
He uses a video-editing analogy: an editor can make every creative call, but you decide whether to publish after reviewing the result. Mapped to agentic shopping:
- Broken model: give AI a needs list and let it decide and buy—hallucinations go unchecked with no veto;
- Trustable model: AI searches, filters and recommends, while you approve line by line.
He then plugs his own Claude connector, LetsFG, which adds live flight and hotel search to Claude, supports split tickets and filtering hotels by specific complaint keywords, and requires email approval before any charge. He notes neither Claude nor his system sees card details (payments run through Revolut), and suggests using it mainly for long trips where human comparison across hundreds of sites can't compete with AI.
More from coding & agent
- Cursor: Malicious Claude Code extension versions sat in marketplace 11 hours, tried SSH backdoor — dl_weekly · 2026-11-09
- Agentic AI can cost 20-30x more than simpler approaches, warns architect David Linthicum — DavidLinthicum · 2026-11-09
- inventory 1.3 released: list packages from 35 package managers in one command — esrtweet · 2026-11-09
- Stanford rewrote CS146S for agents: open-source patches replace one-shot homework — Latent Space · 2026-11-09
- Saudi dev built an AlUla-themed RPG in two days with Claude and Halai — aziz4ai · 2026-11-09
- Indie Dev Launches MCPCore: Build and Host MCP Servers With No Code, Free Tier Included — No_Laugh_996 · 2026-11-09