Integration Infrastructure for Agents and Developers
Ayan Barua, CEO of Ampersand, on infra for AI agent integrations
This marks a shift from selling integration tools to humans, to building infrastructure that both humans and software workers can use directly. Ampersand still has to make setup easy for developers, but now it also has to expose clean actions, schemas, and guardrails that an agent can call on its own inside tools like Cursor. That turns integration UX into both a coding surface and an execution surface.
-
For the developer, the product is ergonomics. They ask for a HubSpot webhook or a YAML manifest in natural language, and the docs MCP server helps generate the integration setup quickly. The job is reducing setup time from weeks of API work to a few prompts and edits.
-
For the agent, the product is a safe tool layer. Ampersand is packaging CRUD style actions across SaaS apps, plus the hard parts MCP does not handle, like auth refresh, field mapping, rate limits, retries, privacy, and tenant isolation. That is what lets an agent act without breaking production workflows.
-
This is also a competitive statement against classic unified APIs. Shallow common models were good for analytics and light integrations, but agentic software needs deeper tenant specific access. That is why Ampersand frames native, field level integrations as more valuable in an AI world than a thin pass through abstraction.
The next step is that more business software will be built with a human operator and an agent sharing the same integration backbone. As that happens, the winning infrastructure layer will look less like a static connector catalog and more like a real time action bus for agents, with strong governance, deep schema awareness, and fast read write loops across systems of record.