Native integrations become basic plumbing
Ayan Barua, CEO of Ampersand, on infra for AI agent integrations
This points to a big shift in where integration stops being a product differentiator and starts becoming basic plumbing. If AI can inspect each customer’s Salesforce fields, Workday objects, or Gong transcripts and map them automatically, the hard part of a native integration shrinks from months of solutions engineering into a short setup task. That makes deep, tenant-specific integrations far easier to ship, and pushes vendors to compete on workflow and outcomes instead of connector coverage.
-
The practical bottleneck in native integrations has always been object and field mapping. Ampersand describes that as the core job, understanding each customer tenant’s custom schema, then connecting it to the app’s own schema. AI changes that because it can infer custom fields and edge cases that older unified APIs flattened away.
-
That weakens the old unified API pitch. Earlier integration layers won by giving developers one simplified model for many systems, but that only worked by ignoring unusual fields and workflows. In an agent world, deeper access matters more because agents need live, detailed context to take actions across CRM, ERP, support, and call data systems.
-
The market does not disappear for middleware, it moves down the stack. MCP gives agents a common way to call tools, but it does not handle auth, retries, rate limits, governance, tenant isolation, or bidirectional sync. That leaves room for infrastructure vendors to make native integrations easy while still owning the operational layer underneath.
The likely endpoint is a market where every serious SaaS product offers deep native connectors for the systems its users already run, while integration platforms become the invisible layer that powers them. As agentic software spreads, the winners will be the companies that can turn customer data from many systems into fast, reliable actions, not the ones that merely expose a long list of connectors.