Read time: ~6 minutes · Series: Product Manager practice · Previous: vs Chat / BI · Next: Quality bar
Business / ops reader?
If your job is review Plan, accept Report, decide in meetings (not evaluating rollout), read Business & Operations introduction—this article is PM / adoption collaboration.
What each role does
sequenceDiagram
participant PM as PM / Business
participant Analyst as Analyst
participant Agent as Agent
participant Eng as Engineering (optional)
PM->>Analyst: Request
Analyst->>Analyst: Focus project
Analyst->>Agent: Start exploration
Agent-->>Analyst: Draft Plan
Analyst-->>PM: Please review Plan
PM->>Analyst: Annotate Plan
Analyst->>Agent: Resolve / revise
Agent-->>Analyst: Incremental Plan edits
Note over Analyst,Agent: Before Execute: Agent won't press or pull formal data
Analyst->>Agent: Press Execute
Agent->>Agent: Run SQL, write Report
Agent-->>Analyst: Report
Analyst-->>PM: Deliver Report
PM->>PM: Meeting delivery
Note over Eng: Connections / MCP infra (parallel to analysis flow)
PMs don't need to write SQL or press Execute (unless you're also the analyst). PMs do need traceable consensus on Plan / Report.
Two moments PMs should join
1. Plan review (before Execute)
This is the definition sign-off window. Confirm:
- Analysis goal matches your decision question
- Denominator, time window, stage definitions match business understanding
- No assumptions you never agreed to
How: Visual mode select text → annotate. Don't only Slack "change denominator"—verbal edits get missed. After analyst Resolve, use View changes; Archive resolved items to keep Plan panel clean.
Don't: Push "just get numbers" while Plan is still Planning; that moves risk to Report.
2. Report acceptance (after Execute)
Check numbers answer your question and limitations are honest (e.g. refunds excluded, short sample).
Annotate wording or add context; definition-level errors need new Plan rerun, not Report text only.
Why Execute must be visibly human
Execute = explicit human authorization: "This Plan version is approved; formal pull may proceed."
- Agent cannot substitute in Planning or chat
- Button on Plan toolbar, not hidden chat command
- Accountability: who pressed Execute on which Plan version
Not distrust of AI—separating exploration from formal external numbers.
vs Slack / meetings
| Channel | Good for |
|---|---|
| Slack / meetings | Requests, priorities, decisions |
| Plan annotations | Definition changes, assumption confirmation |
| Report | External citation, HTML export, meeting attachment |
Project .lantide |
Cross-person Plan / Report / annotations (analysis context; no SQL tabs or data files) |
Handing whole workspace (SQL tabs, connections, optional chat) to a colleague is Workspace export—USER_GUIDE §11.1.0 or Platform: .lantide export/import.
Anti-pattern: Meeting agrees definition, Plan unchanged; or Plan on v3, Slack still cites v1 screenshot.
Two things PMs shouldn't do
- Press Execute for analyst without reading Plan—accountability still on analysis owner
- Skip Plan, demand Report—short-term fast, long-term definition disputes cost more
If only time for Quick exploration, label "informal numbers"—don't put in sign-off flow.