Skip to content

Building with agentic AI (non-technical)

At a glance

You don't need to learn to code. You're learning to work with an AI assistant to get things done — outlines, feedback, drafts, landing pages, research. Think of the AI as a fast intern: it needs a goal, constraints, and your review. This guide covers the mindset, safety basics, and first prompts to try.

Agentic AI in Cursor means the assistant can read your project files, propose or apply edits, and chain multiple steps. Key concepts: Plan mode (think together, review before changes) vs. Agent mode (assistant executes directly). Safety: never paste secrets; verify outputs on money/health/legal; use diffs and undo. Custom slash commands extend the workflow.

This guide covers the product-owner mental model for agentic AI collaboration: defining "done" as acceptance criteria, iteration as the normal workflow (not a failure mode), and the Plan/Agent mode distinction as a trust-gradient control. For deeper technical treatment of how agents work under the hood, see agentic AI.

You are not “learning to code” first. You are learning to collaborate with an AI assistant inside a tool (Cursor) so you can move product ideas forward: clearer specs, better copy, structured research, simple web pages, and prototypes — at a pace that would be painful alone.

1. Mindset: you own the outcome

  • You are the product owner. The assistant is like a fast, tireless intern: it needs a goal, constraints, and boundaries (“do not change X”, “keep this under one page”, “ask before spending money”).
  • Outcomes over syntax. You do not need to memorize commands. You do need to say what “done” looks like (“a bullet list of user stories”, “feedback on this pitch”, “a landing page section in plain HTML”).
  • Iteration is normal. First draft → your edits → “tighten this” → repeat. That is how professionals use these tools.

2. Safety and trust

  • Never paste secrets into chat: passwords, API keys, bank or health data, confidential client material. If it happened by mistake, rotate/revoke the credential.
  • Verify important outputs — especially anything touching money, health, legal, or compliance. The assistant can sound confident and still be wrong.
  • You approve changes. In Agent mode the tool may edit files; use diffs (before/after) and undo when something looks off.

3. Cursor basics (minimal jargon)

Idea What it means for you
Chat You type goals and questions; the assistant replies and can read files you allow.
Project / folder Cursor works best when you open a folder (even a small one with a few notes or a simple site).
Plan vs Agent Plan = think together, review, no (or minimal) file changes until you agree. Agent = the assistant can edit files and run allowed steps to implement what you asked.
New chat Long chats get summarized and lose detail. Start fresh for a new topic and paste a short handoff (see working-with-ai-context.md).

If your workspace uses a [Build] / [Discussion] label in chat, that is just a signal: Build = the turn changed something in the project; Discussion = read-only talk. (Common in advanced setups; optional for you.)

4. First wins (try these before the terminal)

Use your own words; examples below are patterns, not magic phrases.

Goal Example prompt shape
Clarify an idea “Critique this product idea. What are the top 3 risks and 3 questions I should answer next?”
Structure work “Turn these messy notes into a one-page outline with: problem, audience, success metric, next steps.”
Product language “Rewrite this for a non-technical reader. Keep tone friendly and cut jargon.”
User stories “From this paragraph, draft 5 user stories in ‘As a … I want … so that …’ form.”
Landing copy “Draft a hero headline, subhead, and three bullets for [describe product]. I will edit tone.”

Ask for format (“table”, “bullets”, “one paragraph”) so the answer fits how you work.

5. Slash commands (/)

Some workspaces define custom slash commands (e.g. /journal, /session-end). They only exist if that workspace ships .cursor/commands/. In this public repo, human-readable summaries live in ../reference/slash-commands.md — you can still use the same ideas by typing them in plain language (see cursor-triggers-and-prompts.md).

6. When you are ready for “a real project folder”

  1. Open a small folder in Cursor (notes, a static site, a doc repo).
  2. Read the session playbook — terminal tips, GitHub, and how people hand off context between chats.
  3. If you need Git (save versions, share on GitHub), use git-basics-for-non-developers.md.
  4. If you might publish public vs keep private work separate, read public-vs-private-git-strategy.md.
Topic Where
Context and new chats working-with-ai-context.md
Session playbook (deeper) cursor-session-playbook.md
Triggers and phrases cursor-triggers-and-prompts.md
First-week checklist ../checklists/first-week-cursor-non-technical.md
Full catalog ../INDEX.md