← Previous · Film 7 of 10
Who owns what
Each fact's meaning has an owner. Owners publish a public, versioned core; consumers build on it, never on each other's marts.
Topics › Analytics engineering › In the weeds of data crafting
An agent on the team: it drafts and checks, with evidence; people approve
An AI agent can read a dbt project, run it and draft a change in minutes. To be trusted, it needs what a new colleague needs, written down: rules, skills and who approves each step. Least access, a query behind every claim, a test it never weakens, and a reconciliation, a diff and CI before people sign off.
Builds on Who owns what
Watch the film
Press Play, or pick a chapter. Turn on Pause and think to stop for one question at the end of chapters.
Download the videoCaption filesRead the script
Where next?
You've seen the agent's work checked and reviewed. Now review it yourself.
See it in the project
Every card in the film shows a real file from the example dbt project. Open the ones each chapter shows, on GitHub, or run the whole project yourself.
- Written down
AGENTS.mdskills/draft-a-model/SKILL.mdskills/reconcile-and-diff/SKILL.mddocs/process.mdrequirements/ - Least access
AGENTS.md - Evidence
AGENTS.mdseeds/reference/student/learner_identity_decisions.csv - The shortcut
models/intermediate/student/int_learner_timeline.sqltests/reconciliation/_reconciliation__tests.ymlAGENTS.md - Reconcile and diff
skills/reconcile-and-diff/SKILL.mdscripts/tools/diff_against_main.py - Review and ship
scripts/tools/diff_against_main.pyREADME.mdAGENTS.md - The same words, four places
skills/review-metadata/SKILL.md
Think it through
The four Pause and think questions, for a class or a team. Open one to see the answer.
In the series: In the weeds of data crafting
Film 8 of 10.
Next · Film 9 of 10 →
Written once
Every copy of a fact is a chance to drift. One home per fact, everything else generated, and a breaking change arrives as a new version with a date.