Stripe Directory data: merchant playbooks lift agent checkout success from 20/28 to 24/28
jeff_weinstein · x · 2026-09-17
- Stripe shares early data on Directory, its index of agent-buyable businesses, claiming higher agentic conversion, fewer tokens, lower latency, and better reliability.
- Method: one agent, one real browser, one real merchant site and card, buying one item under a spending cap; success only counts on a merchant order confirmation.
- The "advice" is a short merchant-specific checkout playbook (URL sequence, known traps, confirmation page, when to ask a human), built from earlier agents' observations; model, tools, and permissions stay identical.
- Key numbers: across 85 runs (46 unadvised vs 34 advised), the stricter paired comparison of 28 tasks shows success going 20/28 → 24/28 — advice fixed 5 runs and broke 1.
- Stripe's bet: every business will want to welcome well-behaved agents, and Directory is its agentic-commerce infrastructure play.
More from coding & agent
- Dev burns 5B tokens a day on open models as multi-agent swarms emerge as a new scaling axis — xeophon · 2026-09-17
- n8n hit by CVSS 10.0 chain: unauthenticated file read to full RCE, PoC out — evilsocket · 2026-09-17
- free-claude-code: open-source proxy runs Claude Code on NVIDIA's free NIM API — anthara_ai · 2026-09-17
- Templafy launches MCP to turn ChatGPT and Claude chats into branded, compliant Office docs — thetripathi58 · 2026-09-17
- Personal agents should request permissions as needed, not all up front — vikvang1 · 2026-09-17
- GitHub rewrote Copilot's agent runtime into 800K lines of Rust, mostly written by AI agents — martinwoodward · 2026-09-17