A Looker dashboard answers a question once. A weekly review answers it every Monday, and someone has to write the recap. The agent reads the dashboard, drafts a one-page brief, flags movements outside tolerance, and routes it to the named business owner. Looker keeps the LookML model. Dock keeps the brief, the owner, and the sign-off. This sub-essay sits under Dock for data and analytics.
Looker and Snowflake stay the system of record for the raw data. Dock is the system of record for what the agent interprets. Each Dock row carries a pointer back to the platform record, agent identity, decision, reviewer, and timestamp. The agent re-fetches platform data via fresh API reads when it needs current state.
The Dock surface: Dashboard Reviews
| Dashboard | Period | Metric movement | Agent read | Business owner | Status | Looker URL |
|---|---|---|---|---|---|---|
| Revenue by Segment | W21 | Mid-market -8.4% WoW | Two deals slipped to W22; pipeline coverage still 3.1x | Priya N. (RevOps) | Signed off | look.co/dash/rev-seg |
| Activation Funnel | W21 | Day-7 retention -2.1pp | Cohort skew from May 12 trial spike, not product regression | Jordan T. (Growth) | Awaiting owner | look.co/dash/activation |
| Warehouse Spend | W21 | Snowflake credits +14% | Backfill job from data eng; expected and bounded | Casey L. (Data Eng) | Acknowledged | look.co/dash/spend |
Each row points back to the Looker dashboard URL and the underlying Snowflake query hash. The agent identity column (not shown above) records which agent drafted the read, in the pattern described in agent identity.
The workflow
Monday 06:00, the agent hits the Looker API and pulls current values for each tracked dashboard. It compares to last week, the four-week trailing average, and the tolerance band the owner set. For any metric outside band, it drafts a two-sentence read: what moved, and the most likely cause based on segment splits and the Snowflake change log. The draft lands in Dock with the owner tagged. The owner accepts, edits, or rejects with a counter-explanation. The signed brief is the artifact the exec team reads at the Monday review. The agent never edits Looker and never writes back to Snowflake.
Why this matters
Dashboards without owners decay. Owners without drafts skip the review. The agent removes the drafting tax and the dashboard owner keeps the judgment call, which is the part that should not be automated. The audit trail is the same trail you would want for any agent-touched workflow, covered in agent audit and compliance. The same architectural split (platform = data, Dock = interpretation) shows up in Cloud 2.0 for engineering and again in the marketing equivalent, Dock for marketing.
The Looker semantic model already governs what a metric means (Looker docs). Dock governs who agreed with what the metric did this week. Analyst seats are the constrained resource in most BI deployments (Gartner ABI Magic Quadrant context), and shifting the drafting work to an agent with a named reviewer is how that resource gets back to model design and ad-hoc investigation.
Set the agent identity model up the same way you would for any reviewer-of-record workflow (agent identity).
Start a Dock workspace for your dashboard reviews.
FAQ
Does the agent change anything in Looker? No. It reads via the Looker API and writes only to Dock. The dashboard, the LookML, and the underlying Snowflake tables are untouched.
What happens if the owner disagrees with the agent's read? The owner edits or rejects the draft. The original agent read is preserved on the row, alongside the owner's revision, so the audit trail shows both.
How does the agent know which dashboards to review? The Dock workspace lists tracked dashboards with their owners, tolerance bands, and review cadence. The agent only drafts for rows in that list.
Can two agents work on the same review? Yes. Each draft carries the agent identity that produced it. If a second agent revises, the row shows both identities and the timestamps, which is the same pattern Dock uses everywhere.
