Cursor Ask Plan Agent practice exercise

The scenario

Cursor works as a real development partner rather than autocomplete only when you drive it with a structured workflow instead of a single open-ended prompt. The Ask, Plan, Agent loop is how you make a change without breaking working code around it: explore the area first, structure the change before touching anything, then let the agent execute against a plan you have already approved.

The exercise asks you to bring a real, small feature or fix from an actual codebase, something self-contained you could reasonably ship in under an hour. That constraint matters. A feature too large to reason about in one sitting defeats the purpose of practicing the workflow cleanly, and a toy example with no real stakes will not teach you what to watch for when a diff comes back from the agent that technically works but does something you did not ask for.

What you are asked to do

The exercise moves through five steps in strict order:

• Scope a small, self-contained feature or fix and describe it in one or two sentences.
• Use Ask mode with file and folder mentions to explore the relevant area of the codebase and have Cursor explain how it currently works and where your change should go, without editing anything yet.
• Switch to Plan mode and write, or have Cursor draft and you then edit, a step by step plan naming the files touched, the order of changes, and what should not be touched.
• Run Agent mode against the approved plan, review the resulting diff against that plan before accepting it, and note anything the agent got wrong and how you corrected it.
• Turn what worked into a reusable Cursor rule, slash command, or prompt pattern, written so a teammate could apply it to a similar task without the context you built up along the way.

What a strong answer covers

The exercise's own objectives are the standard a strong submission is checked against:

• Is the scoped feature genuinely small and self-contained, the kind of change that could ship in under an hour, rather than something that quietly requires touching half the codebase?
• Does the Ask mode exploration actually happen before any code changes, with a real summary of what was learned, rather than being skipped in favor of jumping straight to execution?
• Does the plan name specific files, a specific order, and what should explicitly not be touched, rather than a vague description of the intended outcome?
• Does the diff review compare what the agent actually did against the approved plan, with any deviation named and corrected, rather than accepting the diff because it looks plausible?
• Is the final reusable rule specific enough that someone without the exercise's context could actually apply it, rather than a personal note only the author would understand?

Skills this exercise practises

Exploring unfamiliar code safely before changing it, instead of asking an agent to guess its way through an area you have not looked at yourself. Structuring a change as an explicit plan the agent executes against, rather than a single freeform request. Reviewing a diff against what was actually approved, catching a plausible-looking change that drifted from the plan. Turning a one-off success into a rule someone else can reuse. These map onto the bootcamp's own curriculum on Cursor interface mastery, structured prompting frameworks, and the Ask, Plan, Agent workflow itself. If you want the equivalent exercise built around Claude Code instead, the Claude Code operating model exercise from Building with Claude Code covers a comparable discipline at a system level rather than a single feature. For a broader look at Cursor itself, see how to build a prototype with Cursor.

Which bootcamp this comes from

This exercise is the practical challenge from Building with Cursor, a one week bootcamp on Builders Camp with two live sessions covering LLM fundamentals for developers, Cursor interface mastery, structured prompting, the Ask, Plan, Agent workflow, and team standards, taught by Tiago Pedro da Costa. Completing the practical challenge counts toward the bootcamp's completion requirement and its certificate, alongside the certification quiz.

Builders Camp runs this bootcamp both live and self-paced, included with the Builders Camp Membership alongside every other bootcamp, track, and masterclass. For a comparison against a different AI coding tool, see how to build a prototype with Windsurf.