Our summary
Joe Herbert at Maia and Matillion wants teams to stop copying slightly different AI integrations into each project. In the first lesson of his three-part series, he builds one connector to TypeSafe’s Jev service, then shares the pipelines that send questions and turn answers into stored data.
Herbert first creates the connector from the API reference and connects a secret store for the access key. He supplies written design notes so Maia understands the answer types. After reviewing a plan, he builds reusable parts for sending requests, processing table rows and unpacking answers, alongside examples and instructions for future builds.
The shared parts are a starting point, not a production-ready service. Herbert lists missing handling for rate limits, uncertain answers, changed response shapes and ownership of the shared component. This vendor-produced lesson requires Maia Foundation and configured infrastructure. JevMade read the written companion and retrieved captions, but did not run or watch the demonstration.
Key takeaways
- Build the connector first and store its access key separately. The runner needs permission to read that secret before the pipeline can work.
- Give the building agent accurate design notes and review its plan. Populate example documentation from real executions rather than invented responses.
- Share a reusable component with explicit inputs. Add rate-limit handling, a review route and clear production ownership before other teams rely on it.