The AI-native case for why platforms still win
I’ve heard some version of this in every conversation for the last two months: “We’re just going to build it ourselves with AI. We don’t think we need an integration platform.”
It’s the hottest objection in enterprise software right now. And it’s going to cost the companies that believe it a fortune.
The pitch sounds great
The promise is seductive: point an AI coding assistant at your integration problem, describe what you need in plain English, and watch it generate working code in minutes. No platform licensing. No partner fees. Just vibes and velocity.
For prototypes? It works. You can vibe-code a point-to-point integration in an afternoon. My team uses AI more aggressively in delivery than anyone in the ecosystem and has decades of experience in delivering hardened systems integration patterns. That’s exactly why I can tell you where this breaks down.
The gap between demo and production
The integration that looks great in a demo has none of the things that keep it alive in production.
Error handling and retry logic. When that API call fails at 2 AM during a maintenance window, vibe-coded integrations don’t have circuit breakers or dead letter queues. They fail silently, and you find out Monday morning when the data is wrong.
Transaction management. When you’re moving an order from ERP to CRM to billing and step two fails — do steps one and three roll back? In a platform, yes. In vibe-coded glue? Partial data in three systems and a reconciliation nightmare.
Security — and this one’s about to get worse. Your integrations carry the most sensitive data in your business — customer records, financial transactions, credentials, PII. Today’s threat landscape is already brutal. Now consider that bad actors will be operating with the kinds of AI models they’ll have in six months. Every unmanaged endpoint, every hardcoded credential, every integration without audit trails becomes an attack surface that scales at machine speed. You don’t secure that with artisanal code. You need a platform with policy-based governance, credential vaults, and access control baked in — not bolted on.
The math nobody does — acceleration toward a big ball of mud
A vibe-coded integration takes an afternoon to build and a career to maintain. Every API change, every schema update, every compliance requirement means someone going into custom code by hand. No abstraction layer. No connector marketplace. No version management.
Multiply that by the 20, 40, 100 integrations a mid-market company runs, and you haven’t innovated — you’ve accelerated toward exactly the outcome that MuleSoft’s founder Ross Mason warned about well over a decade ago: a big ball of mud. Point-to-point spaghetti that nobody can reason about, nobody can audit, and nobody can safely change. Not even Claude Opus 4.6. I wrote about this exact pattern years ago — every innovative breakthrough levies additional burdens on integration strategy, and every generation of IT leaders repeats the same mistake. AI hasn’t broken the cycle — it’s made it possible to create the mess faster than ever.
The instability compounds. Each new vibe-coded connection increases the blast radius of every change in every connected system. You’re not saving money. You’re building a house of cards at 10x speed.
The part that’s actually true
The “just build it” crowd gets one thing right: integration delivery has historically been too slow and too expensive. Specialized integration engineers are essential — you need people who understand error handling, data transformation, and API governance to get it right. But the traditional model of billing those specialists by the hour for 12-week engagements made the economics painful for companies.
The real answer isn’t replacing those specialists with AI-generated glue code. It’s accelerating those specialists with AI so they can get it right faster and cheaper. AI-powered delivery on a real platform. The customer gets enterprise-grade infrastructure — error handling, governance, security, monitoring — at SMB speed and pricing. That’s not a compromise. That’s the best of both worlds.
The bottom line
Vibe coding is a prototype strategy wearing production clothes. AI is transforming how we deliver on platforms, not whether we need them.
The most expensive advice in enterprise integration isn’t “invest in a platform.” It’s “you don’t need one.”