---
title: "Dock + Mode: agent-drafted ad-hoc analyses with attributed audit trail"
excerpt: "An analyst agent drafts the Mode report against Snowflake and dbt models, then writes the interpretation to a Dock review row that a senior analyst signs off before the report is shared."
author: mei
category: Use Cases
date: "2026-05-30"
---

A senior analyst spends the first hour of every ad-hoc request rewriting the same SQL skeleton against the dbt model. An analyst agent can draft that Mode report against the governed Snowflake warehouse. The risk is not the SQL. The risk is the interpretation shipping to a VP without a human reading it. Dock holds the interpretation as a reviewable row, attributed to the agent that wrote it, with a pointer back to the Mode report and a sign-off from the analyst on call.

Mode, Snowflake, and dbt 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 Analysis Review table

| analysis_id | mode_report_url | question | agent_summary | agent | reviewer | status |
|---|---|---|---|---|---|---|
| AN-4471 | mode.com/dock/reports/4471 | Q1 enterprise churn drivers | Three contracts churned. Two cited price during renewal calls. Sample size too small for a trend claim. | analyst-agent-mei | priya@ | approved |
| AN-4472 | mode.com/dock/reports/4472 | Why did NA signups dip in week 18 | Dip is real (down 14% WoW) but isolated to organic. Paid steady. Likely tied to the search console drop flagged Monday. | analyst-agent-mei | priya@ | revisions_requested |
| AN-4473 | mode.com/dock/reports/4473 | Activation rate by onboarding cohort | Cohorts A and B converge by day 14. Cohort C lags 8pp. Recommend a follow-up cut by ICP segment before sharing. | analyst-agent-mei | unassigned | awaiting_review |

Each row links out to the Mode report. The agent_summary is the agent's reading of the chart. The reviewer reads it against the chart and either approves, requests revisions, or rewrites it.

## Workflow: from Mode report to signed-off analysis

A PM asks in Slack why NA signups dipped last week. The analyst agent opens a Mode notebook, queries the dbt model for weekly signups by channel, and confirms the dip is organic-only. It writes the chart, then opens a Dock row pointing at the Mode URL with its summary. The on-call senior analyst opens the report next to the Dock summary, notices the agent did not cross-check the search console drop, and requests a revision. The agent re-reads the search console table, updates the report, and updates the summary. The reviewer approves. Only then does the row mark `approved` and the PM gets the link.

## Why this matters

Mode is the [intelligence layer for the data stack](https://mode.com/), built for SQL, notebooks, and reports. It is not built to hold the agent's read of what the numbers mean, or to gate the read behind a human signature. Fivetran's analysis of [modern data stack workloads](https://www.fivetran.com/blog/modern-data-stack-vs-open-data-infrastructure) notes that agent workflows chain dozens of queries per request, which makes per-request review the only practical gate. Dock is the review surface. It sits next to the warehouse, not inside it.

This is the same pattern across the [data analytics cluster](/blog/dock-for-data-analytics): the BI tool keeps the chart, Dock keeps the interpretation and the signature. It is also the [Cloud 2.0 read for engineering](/blog/cloud-2-0-for-engineering) of how agent work gets reviewed without slowing the human down. The [audit trail](/blog/agent-audit-and-compliance) is the row itself. The [agent identity](/blog/agent-identity) on the row is what makes the sign-off meaningful.

[Start the analyst-review template in your Dock workspace.](/blog/dock-for-data-analytics)

## FAQ

**Does the agent run queries in Snowflake directly or through Mode?**
Through Mode, against the dbt models. Mode handles the connection and query history. Dock holds the interpretation and the signature.

**What if the reviewer disagrees with the agent's reading?**
They request revisions or rewrite the summary. The row keeps both versions and the final approver.

**Can the agent ship a report without a reviewer?**
No. The row stays in `awaiting_review` until a named analyst approves. See [agent audit and compliance](/blog/agent-audit-and-compliance) for the gating model.

**How does this handle stale data?**
The agent re-fetches from the warehouse on each request. The Mode URL stays stable, the underlying query runs fresh.