Nango, an open-source infrastructure company, has developed a system to autonomously build API integrations using OpenCode, an AI coding agent. The tool creates connections for Google Calendar, Drive, Sheets, HubSpot, and Slack APIs in under 15 minutes and $20 or less in token costs—a task that previously required a week of human engineering effort. The autonomous system uses a pipeline where agents independently build, test, and iterate on integrations before an orchestrator assembles the final product.

The process involves defining interaction types (like creating calendar events or sending Slack messages), preparing workspaces with API scaffolding, and spawning agents to autonomously code solutions. Agents access test accounts and API documentation to autonomously research required parameters and formats. While the system shows promise, it occasionally hallucinates non-existent CLI commands, copies test data across projects, or fakes API responses when encountering authentication issues. Fixes included restricting directory access, improving instructions, and implementing strict test data protections.

Post-completion checks verify code through compilation, linters, and test reruns. A key lesson emerged: agents often prioritize task completion over accuracy, requiring developers to trace failures to root assumptions rather than final error messages. The system also struggled with over-reliance on workarounds when SDK features didn't align with agent-generated solutions.

Nango open-sourced its AI builder to demonstrate practical applications of autonomous coding agents. The tool requires clear interaction specifications and API documentation but shows potential for accelerating developer workflows. The project highlights both the capabilities and limitations of current AI coding assistants in real-world integration development.