Single questions are nice, but the real payoff is work that runs on a schedule without you. Your vibe-code agent can write scripts and scheduled jobs that pull data from other systems, keep your Iceberg tables healthy, and react when alerts fire. Each workflow below chains two or more steps. Ask for one, watch it run end to end, then let it repeat.

1. Nightly sync from your store

Most businesses already have data flowing through a store or an app, and retyping it into dashboards is a waste of an evening. The agent sets up a small script that pulls the day’s orders from your WooCommerce API, lands the raw JSON in a MinIO landing bucket, upserts the rows into the orders Iceberg table, and appends a line to a sync log. The next morning your Grafana numbers simply match the store, and nobody exported anything by hand. Try: “Every night at 2am, pull yesterday’s orders from WooCommerce, append them to the orders table, and log how many rows were added.”

2. Morning data-quality report

Bad rows slip in quietly: null customer IDs, duplicate order numbers, negative amounts that break a total nobody checks. The agent writes a scheduled job that runs a handful of DuckDB checks against every key table each morning and sends you a short pass/fail report, with sample offending rows attached so you can see the actual damage. You fix the source once instead of discovering the mess in a board meeting six weeks later. Try: “Each morning at 7am, check the orders table for null customer IDs, duplicate order numbers, and negative amounts, and email me the results.”

3. Weekly compaction and snapshot expiry

Iceberg tables fragment over time: lots of small files make queries slow, and old snapshots grow storage for no benefit. The agent schedules a Sunday job that rewrites small files into larger ones, expires snapshots older than two weeks, and removes orphaned files, then reports the before-and-after sizes. Think of it as vacuuming the whole stack — boring, and essential. Try: “Every Sunday at 3am, compact all Iceberg tables to a 512MB target file size, expire snapshots older than 14 days, and send me the space we saved.”

4. Sales digest to Slack or email

A number you see every day gets acted on; a number you have to go hunt for does not. The agent writes a query bundle for revenue, order count, and average order value, formats the output as plain text, and posts it to Slack or emails it on weekday mornings. No dashboard login, no digging through filters, just the three numbers in the channel where people already talk. If a number looks off, you reply to the message and ask the agent why. Try: “At 8am on weekdays, post yesterday’s revenue, order count, and average order value to the Slack #sales channel.”

5. Self-healing alert response

Your Grafana alert fires “orders data stale” — then what? Usually, someone notices an hour later and starts poking around. The agent can wire that alert to a webhook that wakes it up: it checks when the sync last ran, re-runs the pipeline if it is behind, queries the table to confirm fresh rows landed, and posts a summary of what happened. You wake up to an incident that already closed itself, with a written record of why. Try: “When the orders-stale alert fires, check the last sync time, re-run it if it’s behind, and post what you found.”

6. Mirror Postgres into the lakehouse

Your operational app probably runs on Postgres, but heavy analytics belongs in Iceberg where it cannot slow the app down. The agent builds an incremental mirror that pulls rows changed in the last 15 minutes and upserts them into an Iceberg copy you can join freely with orders and events. Deletes, late-arriving updates, and type quirks get handled explicitly so the mirror stays trustworthy. Once it runs, the join you always wanted is one query away. Try: “Replicate the users and products tables from our Postgres app database into Iceberg every 15 minutes.”

7. Schema drift watch

Upstream systems change their exports without asking you first, and the breakage shows up downstream as failing dashboards. The agent snapshots the schema of every table each day and diffs it against the previous day; a new column, a renamed field, or a type change triggers a warning with the exact diff. You hear about the change the morning it happens, not when a panel fills with red. Try: “Every morning, compare today’s table schemas with yesterday’s and warn me if any column changed name, changed type, or disappeared.”

8. Storage guardrails

Object storage feels infinite until the disk is not. The agent sets up a weekly size report per bucket, warns you at a threshold you choose, and applies MinIO lifecycle rules so raw landing files expire after a set period while curated tables are kept as long as you want. Costs stay boring and predictable, and you get warned while there is still time to do something. Try: “Warn me when any bucket passes 100GB and automatically delete raw files in the landing bucket that are older than 90 days.”

9. One-request onboarding of a new data source

Adding a new source is usually a slog of credentials, buckets, table creation, and a first dashboard — half a day of plumbing per source. Hand the agent the export or API details and it does the whole chain: new bucket in MinIO, historical load, Iceberg table registered in the catalog, a starter Grafana dashboard, and a note in your data catalog. A day of plumbing becomes one conversation with checkpoints. Try: “Onboard our helpdesk export: create a bucket, load the last 12 months into an Iceberg table, and build a dashboard with ticket volume and median response time.”

10. Monthly restore drill

Backups you never test are hopes, not backups. The agent automates a monthly drill: restore the orders table from last month’s snapshot into a temporary table, compare row counts and checksums against the live one, then drop the temp table and report. If the numbers ever fail to match, you find out on a quiet Tuesday morning, not during a real incident at quarter end. Try: “On the first of each month, restore the orders table from the previous month’s snapshot into a temp table, verify the counts match, then clean up and report.”

See it in action

Official walkthroughs from the MinIO and DuckDB teams:

Worth a watch next:

• Migrating Data to Apache Iceberg on MinIO AIStor – From Legacy Lakes to Modern Lake Tables — an end-to-end data pipeline into managed Iceberg tables, the same shape as the sync and mirror workflows above.
• DuckCon #7 – State of the Duck (Hannes Mühleisen, Mark Raasveldt) — the DuckDB project’s own rundown of where the engine and DuckLake are headed, useful context for any scheduled job you build on it.

Workflows are where the agent stops being a chat toy and starts being an operator. Start with the nightly sync or the morning quality report, then add the rest as trust builds. Data Lakehouse (Iceberg + MinIO + DuckDB + Grafana) on OpenSysLab