Demo walkthrough: a PostgreSQL app deployed from a coding agent

TL;DR: This is the written companion to a demo video: a small PostgreSQL app, built and deployed from a coding agent, with the exact prompts and the repo layout. Timestamps, the repo URL and the tracked links are placeholders until the video is recorded. Each step ends with a check you can run yourself.

The end state

Before the steps, the result. A request to the deployed app should return a row from Aiven for PostgreSQL:

curl -s https://<app-host>/notes | head
# [{"id":1,"body":"hello from the agent"}]

If that works, everything below worked. If it does not, jump to the failure table at the end.

Repo layout

notes-app/
  app.py              # FastAPI app, reads DATABASE_URL
  schema.sql          # one table, idempotent
  Containerfile       # builds the app image
  pyproject.toml      # managed with uv
  prompts/            # the exact prompts used in the video
    01-create.md
    02-schema.md
    03-deploy.md

The prompts live in the repo so a viewer (or another agent) can replay them. See deploy a PostgreSQL-backed app to Runtime from your agent for the deploy details this post leans on.

Step list

flowchart LR
    A["00:00 intro"] --> B["01:00 create service"]
    B --> C["03:00 schema"]
    C --> D["05:00 app"]
    D --> E["08:00 deploy"]
    E --> F["10:00 verify"]

All timestamps are placeholders [TO FILL after recording].

The prompts

Prompt 1, create:

Create an Aiven for PostgreSQL service named notes-db in project jay-miller,
cloud do-nyc, on the free plan if available. Tell me the plan first and wait
for my yes. Do not print any connection string.

Prompt 2, schema:

Read DATABASE_URL from the environment. Apply schema.sql with psql using
sslmode=verify-full and ca.pem, then list the tables.

CREATE TABLE IF NOT EXISTS notes (
  id         bigserial PRIMARY KEY,
  body       text NOT NULL,
  created_at timestamptz NOT NULL DEFAULT now()
);

Prompt 3, app and deploy:

Write app.py: FastAPI, psycopg pool, GET /notes and POST /notes, DATABASE_URL
from the environment. Then write a Containerfile and deploy to Aiven Runtime.
Stop and show me the deploy command before running it.

The app itself is deliberately boring:

import os
from contextlib import asynccontextmanager
from fastapi import FastAPI
from psycopg_pool import AsyncConnectionPool

pool = AsyncConnectionPool(os.environ["DATABASE_URL"], open=False)

@asynccontextmanager
async def lifespan(app):
    await pool.open()
    yield
    await pool.close()

app = FastAPI(lifespan=lifespan)

@app.get("/notes")
async def list_notes():
    async with pool.connection() as conn:
        cur = await conn.execute("SELECT id, body FROM notes ORDER BY id")
        return [{"id": r[0], "body": r[1]} for r in await cur.fetchall()]

@app.post("/notes")
async def add_note(body: str):
    async with pool.connection() as conn:
        await conn.execute("INSERT INTO notes (body) VALUES (%s)", (body,))
    return {"ok": True}

Why bother with a written version

An engineer walking through a real build tends to be more convincing than a launch post, and the October tests recorded exactly that for a comparable post. A video is hard for an agent to read, though, so the written companion carries the prompts and the layout in text. It is published on aiven.io/blog and on LinkedIn, each with a tracked link (<tracked-link-blog>, <tracked-link-linkedin>).

When it breaks

What to verify

• [ ] Every timestamp is replaced after the video is cut.
• [ ] The repo URL and both tracked links are real and resolve.
• [ ] The deploy command matches current Aiven Runtime docs [TO VERIFY].
• [ ] psql and the app both use sslmode=verify-full with the project CA.
• [ ] No credential appears in the repo, the prompts folder or the video.
• [ ] The prompts in the post are byte-for-byte the ones shown on screen.