The Future of AI in FinTech Are Not MCPs (medium.com)

Didier Lopes (founder of OpenBB) argues that the future of AI in finance isn’t about proprietary models or brittle connector standards like MCP/FDC3 but about agentic, bot-first workflows that orchestrate intelligence across secure, governed canvases. He contends frontier LLMs (OpenAI-style) win on transfer learning and multimodal reasoning, so most firms should augment rather than rebuild foundation models. The real product becomes intelligent workflows and interfaces that let analysts and agents co-create situational insight — not static apps stitched together with fragile connectors. Technically, Lopes flags critical blockers in current agent protocols (prompt-injection, tool-poisoning, agent-hijacking) that make them unsafe for high-stakes finance, and calls for a reversal in data design: feed LLMs richly tagged, API-first data instead of human-facing PDFs. The proposed “generative canvas” is an ephemeral decision surface that assembles components (geopolitical risk maps, supplier dependency graphs, predictive cost charts) on demand; “vibe coding” is the natural-language orchestration layer for creating reusable tools and directing inference. The implication: FinTech should invest in secure orchestration, data telemetry and governance, and composable workflows that make agents reliable partners — shifting value from owning models to architecting actionable, auditable intelligence.