Evolving Bank API platforms for humans & Agents
summary
Bank APIs at Standard Chartered evolve from portals and sandboxes (scaled to 50 markets) toward machine-consumable, intent-first catalogs because LLMs and agents are nondeterministic while endpoints remain deterministic. Experiments with an MCP-based autonomous payment initiator expose failures: hard errors, missing intent, missing pricing/SLA context and exhausted API limits. Ongoing controlled pilots show APIs must return guided fallbacks, richer intent and next-step guidance, and expose agent-facing metadata while regulators and OpenAPI standards catch up.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 1:11API limits for agents
- 3:01Market forces driving evolution
- 5:12Deterministic APIs vs LLMs
- 6:15Payment API challenges
- 7:42APIs as conversational tools
- 9:10Intent-first platforms
- 12:12API engineering practices
- 15:21Standards for conversational APIs
- 16:44MCP protocol for discovery
- 18:27Regulatory constraints
- 19:53Pilot APIs for agents
- 20:54Explainable and navigable APIs
- 22:40Q&A
- 23:02Closing
related talks
21mPlatforms for Humans and Machines: Engineering for the Age of Agents
31mThe API in 2017
24mThe API Economy's Next Chapter: AI Agents as First-Class Clients
28mOpey, the AI for API Discovery and Future of Fintech
26mAPIs in Agentic world
49mPanel Discussion: Driving the Agentic Shift in Banking Adoption Governance and Scale
35mUnleash Your Inner API Architect: Build Agents That Deliver
33mOptimizing Bank Operations: Design Patterns, NFR, and GenAI Fusion