Can an AI assistant create a database—and can you verify what it saved? I tested the official Prisma Codex plugin on a separate Free Plan database in Paris. The test created a table, saved a fictional 20-minute run, changed its note, and read the updated record back in a separate request. Prisma's announcement connects three layers: Prisma ORM, Prisma Postgres and Prisma Compute. This video separates that broader product promise from the database operations we actually tested. WHAT WORKED • A new isolated Prisma Postgres database was created. • One synthetic record was inserted and edited. • A separate read returned the edited value. • The owner approved database creation, the schema change and the write separately. WHAT THIS DOES NOT PROVE We did not test the ChatGPT interface, a deployed application, sign-in, separation between real users, ORM execution or Compute hosting. This is a bounded Codex-plugin database test, not an application-security audit or performance benchmark. The Free Plan was observed on 9 November 2026; no purchase or public deployment was made. CHAPTERS 00:00 One prompt, three layers 00:21 What we actually tested 00:39 Save, edit, read it back 01:02 Three owner approvals 01:18 What the test does not prove SOURCES Prisma announcement: https://www.prisma.io/blog/one-chatgpt-plugin-to-deploy-it-all Official Codex integration: https://www.prisma.io/docs/ai/tools/codex Prisma MCP documentation: https://www.prisma.io/docs/ai/tools/mcp-server Agent enrollment and approval controls: https://www.prisma.io/docs/prisma-compute/agent-enrollment Prisma Compute: https://www.prisma.io/docs/compute PRODUCTION Original local Blender animation and graphics, AI-assisted cover artwork, original synthesized music and sound effects. Synthetic English narration: Supertonic-3 M1. The miniature workshop illustrates the process; it is not a recording of Prisma's interface. No affiliation with Prisma. This test uses synthetic data only. #Prisma #AICoding #Postgres