Ask vs Plan

Start Here · 3 of 3

Choose between Ask and Plan so you can explore open-ended questions or produce repeatable deliverables with your Agent.

There are two ways to work with your Agent. Ask is exploratory. Plan is structured. Both are equally capable, and you’ll use both — sometimes in the same session.

1. Planning is always present

You saw a glimpse of this in your first session. When you asked your Agent a real question, it laid out a small plan, inspected your data, ran queries, and synthesized the result. That planning is automatic — it happens whether you ask for it or not.

Plan goes a level deeper. Instead of letting your Agent figure out its own approach in the moment, you ask it to write the approach down first, review it together, and save it for next time.

2. Compare Ask and Plan

Use Ask when you’re exploring. Use Plan when you’re producing.

AskExploratory

When

You want to explore a question without committing to where it goes.

Path

Question → answer → follow up → pivot.

Output

Answers in chat. A deliverable at the end if you want one.

Reuse

Convert the session into a Plan after the fact.

PlanStructured

When

You’re creating a report, brief, dashboard—especially if recurring.

Path

Make plan → review → save → execute.

Output

A plan file in Artifacts plus the deliverable it produces.

Reuse

Designed for reuse. Same plan, new data, apples-to-apples comparisons over time.

The Plan path is always the right approach for any recurring deliverable — weekly digests, monthly reports, quarterly reviews — anything where consistent shape over time matters more than any single answer.

3. Start a Plan

Ask your Agent to generate a plan, name the deliverable, and tell it what kind of output you want.

Example
Prompt

“Using the available data, create an analysis plan for a monthly customer experience digest. Identify the top recurring issues, quantify how each has shifted over the past quarter, and surface representative verbatims for each. Reference the relevant tables and columns. Output as a markdown file I can save.”

What comes back is an analysis plan — the approach your Agent intends to take, the data it’ll use (specific tables and columns), the analytical steps in order, and the output spec.

Approach

The research design — how the question will be tackled, including any statistical methods if relevant.

Data sources

Specific tables and columns the Agent intends to query, named explicitly.

Analytical steps

The sequence of analyses, in order. This is what your Agent will execute when you run the plan.

Output specification

The shape of the deliverable — sections, formatting, how verbatims and metrics are presented.

Review the plan before you run it. Push back on anything that’s off: the wrong table, a missing segment, or the wrong approach to a tricky measurement. Your Agent has the training of a seasoned market researcher and statistician; reviewing the plan is the best way to see what it can do, and the cheapest place to catch mistakes.

When you’re satisfied, save the plan file to Artifacts so you can re-run it later. Same plan, new month, comparable output.

4. Use Ask and Plan together

A useful pattern: explore a question with Ask, find something worth tracking, then ask your Agent to convert what you did into a Plan you can re-run on a cadence.

The reverse works too. Plans aren’t rigid — you can deviate mid-execution, follow a thread the data surfaces, and return to the plan when you’re ready. Treat them as living documents, not fixed recipes.

5. Going further

Context engineering

The advanced layer

More sophisticated Plan executions layer in additional context files — each one a guardrail or instruction set that shapes how your Agent works.

AGENT.mdSKILL.mdDESIGN.mdschemasanti-patterns

These are the building blocks of context engineering — the discipline of curating exactly the right inputs so your Agent produces exactly the right outputs. Setup is covered in Setting up your workspace when you’re ready to go deeper.