Read time: ~8 minutes · Role: Product Manager / decision-maker · Next: Follow the reading path below to decide whether it's worth trialing or rolling out to your analytics team.
Questions you may be asking
"We already have ChatGPT / a BI Copilot—why switch to yet another desktop tool?"
"When the Agent runs an analysis and gets it wrong, who's accountable?"
"How is this different from Julius-style tools that turn an uploaded CSV into charts?"
The core question isn't "can AI analyze data" but: can the team be held accountable for a number, review it, and re-run it—instead of reading a chat summary aloud in the weekly meeting.
What Lantide Data is (PM view)
In one sentence: a local data-analysis IDE that offers a governable, built-in or external Agent way of working—it turns analysis into a reviewable Plan / Report / SQL, and only a human can press Approve & Execute to kick off a formal analysis.
It is not:
- A general chat model (no workspace, Plan state, or cached lineage)
- An "upload a CSV, get a chart" chat-analysis tool (definitions stay inside the conversation and are hard to treat as formal deliverables)
- A repo-editing coding agent (which ignores the Plan, Report, and Execute boundaries)
It sits between "quick Q&A AI analysis" and "code-editing agents"—it deliberately exposes SQL, makes the Plan annotatable, and keeps Execute in human hands, trading governability for deliverables that stakeholders can sign off on.
Why it's worth recommending to your analytics team
1. Clear accountability boundaries
Execute = an explicit human authorization: "I approve this version of the Plan; it's cleared to run for real." The Agent can't pretend you've already approved during the Planning stage or inside a conversation. In the weekly meeting you can ask: who signed off on this definition, and under what assumptions—rather than "that's how the AI happened to calculate it at the time."
2. Collaboration happens on the document, not in Slack back-and-forth
Annotations on the Plan are bound to the original text; once Resolved, the Agent makes incremental edits. Consensus is written into Markdown, not verbally agreed and then forgotten.
3. Analysis quality is judged by content, not form
Whether an analysis is ready to take to a meeting for sign-off comes down to: can the Plan be executed through to a Report, does the Report have concrete numbers and limitations, does the funnel have stage definitions and denominators—not cosmetic metrics like how elegantly the pipeline is decomposed. Multi-step materialization and Source Run can speed things up and aid debugging, but they cannot replace a reviewable definition and deliverable.
4. Memory is governable, not a black-box accumulation
Agent proposals go into Queued Knowledge, and are only written into User / Project knowledge after you approve them. Sensitive projects can turn this off; conversations are not fed into a vector store by default.
5. An external Agent's permissions and analysis approval are two separate gates
In Agent Integration, Observe, Execute, and Admin are the connection's capability ceiling; a Plan's Planning, Executing, and Executed states are the analysis lifecycle. Granting an external Agent Execute or Admin does not equal approving a particular version of the Plan—a formal analysis still requires a human to press Approve & Execute. For governance choices, see External Agent Integration.
Who should use it, and who might not
| A good fit | Not a great fit |
|---|---|
| Analysis collaboration that needs aligned definitions and sign-off-ready deliverables | One-off charts only, with no need for SQL evidence |
| Local CSV / Excel scenarios | Heavy cloud warehouse + enterprise ETL scheduling by default |
| Wanting the Agent to explore while humans keep Execute and annotation | Expecting the Agent to produce reports fully automatically with no human review |
| Small teams treating "Plan as spec, Report as deliverable" | Only needing one-shot NL2SQL Q&A |
→ Detailed comparison: How it differs from ChatGPT / Julius-style tools
Suggested reading path (onboarding → practice → advanced)
Onboarding: make the adoption decision (~15 minutes)
| # | Article | Overview | Status |
|---|---|---|---|
| 1 | Who should adopt, who shouldn't | Team size, data types, collaboration needs, and Lantide fit | Available |
| 2 | How it differs from ChatGPT / Julius-style tools | Chat analysis vs SQL-first; the governability trade-off | Available |
Practice: how the team adopts it (~25 minutes)
| # | Article | Overview | Status |
|---|---|---|---|
| 3 | Cross-role collaboration: how PMs use it | Division of labor between humans and the Agent; why Execute must be observable | Available |
| 4 | The quality bar: review before Execute | Numbers, denominators, limitations; form metrics can't replace content | Available |
| 5 | Don't confuse Agent Memory with the Plan | Queued Knowledge approval; compliance and off-switch strategy | Available |
| — | Follow along | USER_GUIDE §2, or have an analyst give you a demo | Available |
Advanced: outreach and adoption (~15 minutes)
| # | Article | Overview | Status |
|---|---|---|---|
| 6 | The five-step team rollout | Pick a real small project, run Plan→Report end to end, observe the Execute boundary | Available |
| 7 | Elevator-pitch material | A four-part script: category, differentiation, scenarios, boundaries | Available |
Design depth (optional)
| Resource | When to read |
|---|---|
| 0. Series guide and product positioning | Market gap, the Julius comparison, product category positioning |
| 1. Data-analysis workflow in the Agent era | Full text on the Execute and Plan/Report design |
| 2. Governable Agent Memory | Queued Knowledge governance |
| External Agent Integration | Evaluating an external Agent's scope, mode, Activity, and revocation |
Trial adoption: how to validate it
- Pick a real small project with a contested definition (it conveys the value better than demo data)
- Have an analyst follow USER_GUIDE §2 to run Plan → Execute → Report end to end
- Deliberately leave one Plan annotation and check whether only the referenced section changes
- Confirm that only a human presses Execute—the Agent won't run ahead during the Planning stage
- Assess: can the Report be attached to the weekly meeting, instead of hand-rewriting the logic into another slide deck
Other roles
- Analyst introduction — day-to-day analysis workflow and SQL-first
- Business & Operations introduction — reviewing the Plan, accepting the Report (second pilot wave)
- Platform Admin (Data Engineering) introduction — connections and workspace migration (when you have engineering support)
- Learn Hub