Manik Surtani is Head of Open Source at Block.
Discover top fintech news and events!
Subscribe to FinTech Weekly's newsletter
Read by executives at JP Morgan, Coinbase, Blackrock, Klarna and more
In 2025, AI agents transformed how fintechs operate internally, automating complex workflows and coordinating across tools with minimal human direction. In 2026, we'll see more and more of these agentic features offered directly to customers. But the industry faces a choice. Today's financial technology ecosystem is deeply fragmented. Every payment processor, lender, bank, and platform has its own data formats and APIs. Customers can either get agents that work only within isolated systems, or we can collectively move toward open standards that enable agents to operate across a broader set of financial contexts.
Earlier this month, Block, Anthropic, and OpenAI, in partnership with the Linux Foundation, announced the formation of the Agentic AI Foundation (AAIF), bringing together contributions from their respective companies, with support from other AI leaders, to establish open standards for agentic AI. While early, this represents a meaningful step toward improving interoperability in financial technology. If the industry embraces this direction, we can build an ecosystem where agents can learn from richer data, access harmonized interfaces, and deliver benefits that compound rather than fragment. If we don't, we risk recreating the same siloed architecture that has slowed innovation for decades, only this time with more powerful technology.
The Limits of Agentic AI in a Silo
Fintech has historically grown through proprietary stacks. That model worked in the past, but agentic AI exposes its limitations. Agents need consistent access to context, action surfaces, and signals from multiple systems.
When every institution structures transactions, identities, risk indicators, and merchant profiles differently, agentic AI runs into serious obstacles. Fragmented data undermines agents' ability to reason or take confident action. Integration friction slows deployment and increases engineering costs. Vendor lock-in forces companies to choose less effective tools simply because they fit the existing architecture, or worse still, create their own silos which just compounds the problem.
Agentic AI succeeds when it can observe, decide, and act across connected systems. Siloed environments weaken all three capabilities.
Why Open Standards Change Everything
Open standards (shared schemas, definitions, and protocols) do far more than simplify integration. They create the foundation for scalable and interoperable agentic behavior.
Before agents can reason across systems or act on behalf of users, those systems must speak the same language. Consider the Model Context Protocol (MCP), an open standard that gives AI systems the ability to interact with real-world tools and data. In just about a year, MCP has seen growing adoption across industries, including fintech and commerce companies. Block built the first reference implementation for MCP with goose and was an early contributor to the protocol itself. Stripe built MCP support to let agents access payment data, create checkout sessions, and manage subscriptions. Square released MCP servers for its payments, catalog, and customer APIs. Shopify launched MCP integrations for its commerce platform. These examples illustrate genuine market interest in interoperability.
With interoperable protocols, agents can interpret data with greater contextual understanding. Fragmentation, by contrast, limits the quality of signals on which agents rely.
Contrast this with open banking. Open banking has taken years to progress globally (especially in the U.S.) because it required institutions to do the heavy lifting: building new APIs, ensuring compliance, coordinating across regulators. Progress depended on regulatory pressure, and even then, adoption has been slow and uneven. In both cases, customers benefit from better interoperability. With agentic AI, companies may have an additional incentive: agents can help bridge or translate between systems, lowering integration burdens and making open standards commercially attractive rather than solely compliance-driven.
The next generation of agentic AI will consist of specialized agents that collaborate. One agent may excel at document classification, another at fraud detection, another at cash flow forecasting. Predictable interfaces and shared protocols can help these agents discover services, delegate tasks, and orchestrate workflows without brittle custom code.
Once agents can move fluidly across financial platforms, the real power of interoperability becomes clear. Right now, every financial service operates in isolation. Your payroll system doesn't talk to your business banking app. Your expense management tool can't coordinate with your accounting software. Your payment processor has no visibility into your cash flow forecasting. With open standards, agents can orchestrate across all of these. They can reconcile expenses automatically by pulling data from your corporate card, matching it against invoices in your accounting system, and updating your budget forecasts in real time. They can coordinate payment timing across multiple platforms, ensuring you pay vendors when cash flow is strong and defer when it's tight. They can connect underwriting data from one platform to risk assessment on another, so you're not filling out the same information repeatedly. The value lies in connecting systems that were not originally designed to interoperate.
Smaller fintechs benefit too. Open standards level the playing field by allowing new entrants to connect their agents to banks and processors without expensive engineering projects. They can compete on insight and experience rather than integration budget.
Build the Rails, Not the Walls
The next decade of fintech will be defined by companies that understand that agentic AI is not a single product. It is a platform for reasoning, action, and collaboration across systems. Platforms only scale when the industry agrees on the rails they run on.
AAIF represents an important first step, but it's only the beginning. To unlock the full potential of agentic AI, fintech needs to get involved. We need open data schemas specifically designed for financial primitives: merchants, transactions, identities, risk signals, and payment flows. Some commerce and payment protocols already exist and more are being proposed, but they still need industry-wide buy-in and collaboration to become true standards rather than isolated implementations. We need shared safety and governance frameworks so trust can scale alongside innovation. And we need active participation from fintech leaders in industry groups that define and maintain these standards, not just passive observation.
This does not mean giving up differentiation. The strongest companies will differentiate in experience, risk management, and intelligence, not in proprietary plumbing. The history of the internet shows that strong infrastructure can expand opportunity rather than reduce it. Agentic AI offers a chance to do it again.
About the author
Manik Surtani is the Head of Open Source at Block, Inc. At Block, Manik has previously led engineering teams at Square and Cash App. Before joining Block, Manik was Staff Engineer at Red Hat. He was the founder and lead engineer of the Infinispan project and platform architect on the JBoss Data Grid. Manik has a background in AI, distributed and fault tolerant systems, and performance tuning of JVMs. Manik is a strong proponent of open source development methodologies, ethos, and collaborative processes, and has been involved in open source since his first forays into computing.
