One stat tucked away in a recent Gartner® report just told you exactly where your AI strategy lives or dies.
36% of software engineering leaders say strengthening integration is the single most important action their organization can take to deliver on AI priorities. That’s one in three engineering leaders. They didn’t say model selection or agent frameworks. They said integration.
Here’s why. AI has done more than level the playing field. It’s obliterated it. The old ways of doing things have become invalid seemingly overnight. Every major software purchase right now feels like it could be a ticket on the Titanic. We’ve already spotted the iceberg. The question is whether your platform can turn fast enough.
When speed is everything, your software either enables it or kills it. The integration layer is where you find out if you’re on the Titanic or a speedboat.
When you’re evaluating iPaaS vendors, here are the five questions to ask, the responses that should give you confidence, and the ones that should send you running.
Question 01
Does it support AI agents end-to-end, or just at the edges?
Every iPaaS vendor will tell you they support AI. It’s a claim so ubiquitous it’s become meaningless. The question is whether their platform can actually support the full lifecycle of an agentic workflow inside an enterprise environment.
There’s a significant difference between a platform that can call an LLM and one that can coordinate multiple agents across systems, execute complex business logic deterministically, handle autonomous task planning, and maintain full visibility into what every agent is doing at every step. Most platforms can do the former. Far fewer can do the latter.
AI-ready data is part of this equation too. Agents are only as useful as the context they can access. If your platform can’t transform unstructured data into formats agents can actually work with, you’re building on a foundation that will constrain you before you’ve started.
"We have a native integration with OpenAI and Claude…you can call any model directly from your workflows."
"Our AI features are built on top of our existing connector framework. It's the same platform you already know."
"We support AI agents through our partner ecosystem. There are several third-party options available."
"We support both deterministic agent workflows where you define the logic, and fully autonomous workflows where the agent plans and executes tasks dynamically."
"Our platform automatically chunks and embeds unstructured data into vector tables so agents can access it without any upstream data preparation work."
"You can coordinate multiple agents across systems within a single orchestration layer, with full visibility into what each agent is doing and why."
Question 02
How does it handle MCP governance at scale?
MCP (Model Context Protocol) went from a novel specification to becoming the standard for how AI agents connect to enterprise systems and data in under a year. If a vendor isn’t talking about MCP with some fluency, that’s already a red flag in and of itself.
MCP support became table stakes overnight, but the more important question is what sits around it. Ungoverned MCP sprawl — agents connecting to enterprise systems without proper access controls, authentication, or audit trails — is how you create security exposure at exactly the moment you’re trying to scale AI across the business.
When talking to a vendor, demand specifics. A vendor that can describe their MCP gateway, their authentication model, and their observability capabilities in detail has actually built something. A vendor that talks about MCP support without being able to answer these questions has bolted a label onto something that isn’t ready.
"We support MCP! You can connect your agents to any of our connectors through it."
"Governance and access controls are on our roadmap for H2."
"Each team manages their own MCP server configurations."
"We operate a centralized MCP gateway. That means every tool call is logged with agent identity, what was accessed, and what happened as a result."
"Each agent gets its own OAuth credentials with role-based access. No shared API keys, no agent can access what it isn't explicitly permitted to."
"If an agent attempts a high-risk action or hits a low-confidence decision, the platform routes it to human review before execution."
Question 03
How coherent is the underlying architecture?
This is the question most buyers forget to ask, and the one that tends to cause the most pain 18 months into a deployment.
Integration platforms have been built in fundamentally different ways. Some have been developed as coherent, purpose-built architectures over time. Others have expanded through acquisitions, adding capabilities that weren’t originally designed to work together. Neither path is automatically disqualifying. But the architectural history of a platform shows up in specific, practical ways: inconsistent security models across modules, different governance behavior in different parts of the product, release cycles that don’t align, upgrade paths that require more manual work than they should.
You won’t find this in a product demo, and you don’t want to find it a year later. You find it by asking direct questions about how the platform was built and watching closely for where the answers get vague.
"We've made several strategic acquisitions over the past few years that have really expanded our capability set."
"Those two modules are on different release cycles but we're working on unifying them."
"Governance works slightly differently in that part of the platform but the team adapts pretty quickly."
"The platform was built on a single architecture from day one so every capability, including our AI features, runs on the same runtime and governance model."
"Security policy is enforced consistently across every part of the platform. There's no difference in how we handle access control in integration versus orchestration versus agentic workflows."
"We can show you exactly how a workflow that spans data transformation, agent coordination, and external API calls behaves end-to-end in a live environment."
Question 04
What does total cost of ownership actually look like under real-life demand?
iPaaS pricing is more complex than it appears at the point of selection. Most platforms use some form of consumption-based pricing — message volumes, API calls, data transfer, environments — and the gap between what a platform costs in a proof of concept and what it costs at production scale can be significant.
Agentic workflows add a new dimension to this problem. Every LLM call, tool retry, agent loop, and multiagent handoff generates cost. In complex deployments, those costs compound in ways that are genuinely difficult to predict without modeling them against a real demand profile upfront.
The right approach is to do that modeling before you select a vendor, not after you’ve signed. Give vendors your projected workload in detail and ask them to map it to their pricing model explicitly. Any vendor unwilling or unable to do that is telling you something important.
"Pricing is based on message volume. We can give you an estimate based on your current usage."
"Most customers start small and scale up as they need to. The platform grows with you."
"Agent workflows are priced separately but it's very competitive. We can walk you through it after you sign."
"Give us your projected message volumes, API call frequency, environment requirements, and expected agent workflow activity and we'll model the TCO out 18 months before you commit to anything."
"Our pricing includes volume tiers and caps. Consumption-based pricing without guardrails is a commercial risk we don't think our customers should carry."
"Here's exactly how agentic workflows are priced. Every tool retry, every agent loop, every multiagent handoff. No surprises at scale."
Question 05
Is it built for your team?
Integration is a team sport.
Integration platforms serve genuinely different users, and the best platform for one team can be actively wrong for another. Integration specialists need advanced data transformation, robust orchestration, and enterprise-grade operational controls. Application developers need extensible, code-centric environments with version control and CI/CD integration. Business technologists need intuitive low-code tooling with AI-assisted flow generation.
A platform optimized for one of these personas creates friction for the others. And in most enterprises, all three personas exist and all three will touch the platform. You need to broaden the question from “can our team use this?” to “can every person who will touch this platform work effectively in it?”
Map your actual delivery personas before you shortlist. It’s a straightforward exercise that eliminates a significant amount of post-selection regret.
"The platform is really designed for technical users but we have some low-code features for business teams."
"Most of our customers have a central integration team that manages everything. It works really well."
"Business users can build their own workflows. It just takes a bit of training to get started."
"Tell us who's building, who's operating, and who's consuming. We'll show you what the platform looks like for each of those personas specifically."
"Our platform supports integration specialists, application developers, and business technologists with purpose-built tooling for each. They're not all using the same interface."
"We can put the platform in front of your actual team during the evaluation, not just your architects."
These questions get you to a shortlist. The report takes you further.
These five questions will get you to a meaningful shortlist. But what you really want is the vendor-specific detail — how specific platforms actually perform against each of these criteria, where they’re strong, where they fall short, and what to demand in proof-of-concept demonstrations.
Gartner recently published a guide specifically for enterprises navigating this decision. It goes deep on evaluation frameworks, deployment models, and AI readiness criteria, and it names names.
Gartner® Report
How to Choose the Best-Fit Integration Platform as a Service Vendor
Evaluation frameworks, deployment models, AI readiness criteria — and vendor-specific findings.
Access the report →Gartner® Disclaimer
GARTNER is a registered trademark and service mark of Gartner, Inc. and/or its affiliates in the U.S. and internationally and is used herein with permission. All rights reserved.
Gartner does not endorse any vendor, product or service depicted in its research publications, and does not advise technology users to select only those vendors with the highest ratings or other designation. Gartner research publications consist of the opinions of Gartner’s research organization and should not be construed as statements of fact. Gartner disclaims all warranties, expressed or implied, with respect to this research, including any warranties of merchantability or fitness for a particular purpose.
Gartner, How to Choose the Best-Fit Integration Platform as a Service Vendor, Shrey Pasricha, Keith Guttridge, Andrew Humphreys, 6 April 2026