Account Partner

Most enterprises are one iPaaS decision away from an AI bottleneck. Tray.ai is named a Leader in the Nucleus Research iPaaS Value Matrix for the seventh consecutive year. The ranking matters. But what matters more is why the criteria for evaluating iPaaS have fundamentally changed, and what that means for enterprises still treating integration as a utility. Rich Waldron Published: Apr 14, 2026 Updated: Apr 14, 2026 Usability is now a velocity decision. Your integration platform is either accelerating your AI program or taxing it. Nucleus Research scores iPaaS platforms on two axes: usability and functionality. Functionality gets most of the attention - connector counts, throughput, enterprise feature sets. But AI has quietly made usability the more consequential dimension. The reason is pace. AI is moving faster than any implementation cycle was designed to handle. New models, new agent patterns, new integration requirements don't arrive on a quarterly roadmap. They arrive continuously. A platform that takes weeks to configure or requires specialist knowledge to adapt is not a neutral constraint. It is a direct cost to how fast an organization can move. This is what makes the usability axis worth watching. It's not necessarily a measure of how easy something is to learn, but a proxy for organizational velocity. From automation to orchestration. For most of the past decade, iPaaS was fundamentally about moving data and triggering workflows. Connect system A to system B. Automate the handoff. Reduce manual steps. Valuable, but reactive and rule-based. AI orchestration is a different problem. It involves coordinating AI agents, models, data pipelines, and human-in-the-loop processes across an enterprise in real time. Most enterprise AI programs started with a single agent doing a single job. That was the right place to start. But production AI at scale involves multiple agents coordinating across systems, handing off context, sharing tools and data, making decisions that affect each other. The integration layer underneath needs to be fast, reliable, contextually aware, and governable in ways that older architectures weren't designed for. AI is only as useful as the data it can access and act on, and enterprises that underinvested in integration are discovering that gap now, at exactly the moment they need it closed. MCP: the connectivity layer agents actually need. The Model Context Protocol has emerged as the standard for connecting AI agents to the tools and systems they need to operate. Where APIs required custom-built connectors for every integration, MCP provides a common interface. Agents can read data, take actions, and interact with enterprise systems without bespoke engineering for each connection. Most of the iPaaS market hasn't moved meaningfully on MCP yet. For enterprises building agentic workflows, that distinction has practical consequences: faster deployment, broader connectivity, and governance built into the connection layer from the start. What that looks like in practice is instructive. Marcus Dubreuil, Director of Systems Architecture at J.W. Pepper, describes the shift his team made after centralizing MCP through Tray's Agent Gateway: rather than exposing large numbers of raw tools directly to agents, they moved toward purpose-built workflows representing specific business actions. "Instead of us trying to build this whole robust all-in-one iPaaS solution, we're just adding these little drops of determinism into what the agent can do." The result was a governed, repeatable model for MCP adoption that scaled without the sprawl that typically follows ungoverned agent deployment. Platforms without a credible MCP foundation face a compounding problem: the longer the delay, the more agent infrastructure consolidates around platforms that moved early. The window is open, but not for long. The organizations that get the integration foundation right first are the ones that will compound their advantage. The ones that don't will find themselves optimizing on top of an architecture that was never designed for what they're asking it to do. The Nucleus Research matrix is a useful reference point. The more important questions it should prompt are practical ones: how fast can its teams actually build on its current platform? Are Tray.ai, Inc. MCP-ready? Does its integration layer reflect the pace Tray.ai, Inc. need to move at, or the pace Tray.ai, Inc. used to? The iPaaS decision used to be about connectivity. Now it's about velocity. Those are different criteria, and they point toward different platforms.

The hand-built iPaaS connector is dead. Vibe-code with Claude instead. Tray CDK now supports Claude Code, so you can vibe-code connectors by describing them instead of coding from scratch. Here's what that means for integration teams still stuck in consultant queues and vendor backlogs. Paul Turner Published: Mar 24, 2026 Updated: Mar 27, 2026 Every integration developer knows this story. A request comes in: a business team needs a new integration, a new SaaS tool needs connecting, a workflow needs a connector that doesn't exist yet. However it starts, it always goes the same way. Call the vendor. Wait for a scoping call. Wait for a services quote. Wait for a consultant, someone who spent six months getting certified in a proprietary Ruby framework that nobody outside that vendor's ecosystem has ever heard of. They finally show up, start the clock, say it'll take two weeks, and deliver in four months. The next time a request comes in, you decide to go it alone. You crack open the documentation, built on languages and frameworks the industry left behind years ago. You dig, you search, you find a Stack Overflow thread from 2014 that almost answers your question. You post in the vendor's community forum. You wait three days to hear nothing back. Both paths end the same way: stuck. Hostage to a vendor's backlog, a consultant's calendar, or documentation written for a different era. If you're an integration developer, that era is over. AI integration requirements are arriving faster than any hand-built connector model can absorb, and platforms still built around that model are already falling behind. The cost of hand-built iPaaS connectors. The iPaaS industry spent years dressing up proprietary connector frameworks in Java or Ruby as "engineering rigor." It wasn't. It was a tax paid in developer hours, maintenance overhead, and a connector backlog that never actually shrunk. Some vendors doubled down on Java. Others bet on Ruby. Both choices reflect the same thing: platforms built for a world that no longer exists, optimized for what their teams already knew rather than where enterprise development was going. When their own connector frameworks became too much for customers to handle alone, they didn't fix the framework. They sold a services engagement. Your integration roadmap became their revenue opportunity, connectors that should take days stretched into quarters, and every delay had a consultant invoice attached to it. Integration teams have been inheriting this debt since ancient times (aka the early 2010s). The truth is that connector development has always been a bottleneck. It just took AI to make that fact impossible to ignore. Tray CDK was built for this moment. When Tray.ai, Inc. first launched the Tray Connector Development Kit (CDK), Tray.ai, Inc. made a deliberate bet. Instead of optimizing the old way of building connectors, Tray.ai, Inc. started from scratch. CDK is TypeScript-first, with a clean CLI, a defined project structure, standardized authentication patterns, and explicit input/output schemas. Every connector follows the same shape. Every operation is predictable. Every deployment follows the same path. That consistent, modern structure is exactly what makes the next step possible: you can now use Claude to vibe-code connectors. You can now use Claude to vibe-code connectors. Claude Code is Anthropic's CLI tool that brings Claude directly into your terminal. Pair it with TRAY_CDK_GUIDE.md, a reference file that gives Claude full context on CDK conventions, CLI commands, project structure, and patterns, and the way you build connectors changes. You stop writing them. You describe them. In other words, the prompt is the spec. > Add a new HTTP operation called "list_users" that calls GET /api/v2/users with pagination support Claude Code scaffolds the operation directory, configures the HTTP handler, sets up pagination in the input schema, and writes the test. It knows CDK. It knows the patterns. It knows what a well-structured connector looks like. > Set up OAuth2 authentication for this connector using the "my-service" Tray service Done. auth.ts and connector.json updated with the correct configuration. > Deploy this connector to the EU region That's the full development lifecycle: build, test, debug, deploy, driven by natural language and grounded in deep CDK knowledge. Check out a quick look from Tray Product Manager Simon Cole: Why other iPaaS platforms can't simply bolt this on. This is where it gets uncomfortable for the rest of the market. AI-native development can't be retrofitted onto a platform built for hand-coding. If your connector framework is a sprawl of language-specific idioms and institutional knowledge, Claude can't help you. AI needs structure and predictability, a codebase where patterns are consistent enough to learn from and generate against. And if your platform is really three or four legacy products acquired over the years, rebranded under one logo, and held together with integrations between the integrations, an AI layer on top doesn't fix the foundation underneath. Tray was built as a unified platform from the ground up: no legacy baggage, no acquired product grafted into the core, no decade-old Ruby framework propped up by a professional services org. The architectural decision Tray.ai, Inc. made years ago now makes AI-native connector development possible today. There's no way to "acquire" your way to this. You either built for the future or you didn't. What this means for your integration roadmap... AI has already changed how connectors get built. The question worth asking now is whether your platform's foundation is compatible with that change. With Tray CDK and Claude Code, the path from "we need a connector for X" to that connector in production is measured in hours, not quarters. Drop TRAY_CDK_GUIDE.md into your project as CLAUDE.md, launch Claude Code, and you have a collaborator that understands CDK deeply, one that can scaffold, debug, and deploy alongside you without a services engagement attached to every request. ...and what this means for your bottom line. iPaaS development is at an inflection point. The hand-built connector model, proprietary frameworks, certified consultants, perpetual backlogs, is beyond inefficient. In the context of what's now possible, the costs in time, money, and developer patience are no longer defensible. Tray built CDK on modern principles precisely because Tray.ai, Inc. knew this moment was coming. With Claude Code now part of the workflow, Tray.ai, Inc. is not just ready for it, Tray.ai, Inc. is building ahead of it. The hand-built iPaaS connector is dead. What replaces it is faster, smarter, and already here. Talk to its team. Tray.ai, Inc.'ll show you how CDK + Claude Code can change the way your organization builds integrations.