Bill Raymond Cambermast logo

AI coding

AI Coding Prompt Guide

Learn how to chat with an AI coder successfully, when to start in plan mode versus a focused chat, and how to use Git and GitHub prompts to finish the work cleanly.

Printable PDF version coming soon
AI Coding Prompt Guide featured image.
Start here

Use the right mode before you write the prompt

You do not need to talk like a developer to get useful results from an AI coder. Start by picking the right kind of conversation, then say what outcome you want and whether you also want help with Git or GitHub.

1

Plan mode

Ask the AI to create a Product Requirements Document (PRD) before it edits so you can review scope, tradeoffs, and implementation steps.

Best for

  • New projects or major changes
  • Redesigns or unclear requirements
  • Multi-file work with dependencies
Tip: Use for new projects or big changes.
2

Focused chat

Keep the chat scoped to one page, feature, or product area. Related tweaks can share a session — just avoid mixing unrelated areas.

Best for

  • Website or app updates in one area
  • Bug fixes or wording changes
  • Commit, sync, or verify tasks
Tip: Use for focused changes in one area.

Two ways to work with an AI coder

The most important choice is not perfect wording. It is whether you need the AI to create a Product Requirements Document (PRD) first or jump into a focused chat scoped to one area that gets the work done.

1

Plan mode

Use first for projects and larger changes

PRD first

Ask the AI to create a Product Requirements Document (PRD) before it edits so you can review scope, tradeoffs, and implementation steps.

Use this when

  • New projects or major changes
  • Redesigns or unclear requirements
  • Multi-file work with dependencies

What to ask for

  • A PRD or implementation plan
  • Tradeoffs and missing decisions
  • Acceptance criteria, testing, and rollout steps
Best practice: Use for new projects or big changes.

What to expect

You should get a PRD, open questions, and a clearer implementation path before code changes begin.

Why this works

Planning mode reduces rework, surfaces hidden decisions early, and keeps bigger efforts from turning into messy chats.

When not to use it

Do not use plan mode for a one-line copy tweak, a small app fix, or normal day-to-day commit and sync work.

2

Focused chat

Use for day-to-day work and focused changes

Do the work

Keep the chat scoped to one page, feature, or product area. Related tweaks can share a session — just avoid mixing unrelated areas.

Use this when

  • Website or app updates in one area
  • Bug fixes or wording changes
  • Commit, sync, or verify tasks

What to ask for

  • The exact change you want
  • Any constraints that matter
  • A commit, GitHub sync, or build verification if needed
Best practice: Use for focused changes in one area.

What to expect

You should get a focused implementation, a short summary of what changed, and any requested verification at the end.

Why this works

Focused chats reduce context drift, speed up execution, and make it easier to review whether the change is correct.

When not to use it

Do not use a focused chat for new projects, redesigns, or work with many moving parts.

Role differences

These examples show how a developer and an AI developer approach the same situation differently. The developer frames the engineering task, and the AI developer helps scope, execute, document, and verify the work.

SituationDeveloperAI Developer (prompting)
Start a new project or major featureCollaborate with the product manager to create a Product Requirements Document (PRD)
Use plan mode to collaborate on scope, draft the Product Requirements Document (PRD), surface open decisions, and outline implementation, testing, and rollout before coding starts.
Make a small page or app changeMake one targeted change in place
Implement the requested change, stay within the existing design or code patterns, and summarize what changed when the task is complete.
Commit the workCreate a clean Git checkpoint
Prepare the completed work as a clean commit with an appropriate message.
Sync with GitHubUpdate the remote branch safely
Push the work to the remote branch and handle any branch reconciliation safely.
Commit, sync, and verifyShip the change and verify CI
Complete the delivery workflow by committing the work, syncing it to GitHub, monitoring CI, and reporting the final result.
Ask for a stronger promptClarify the spec before implementation
Translate a rough request into a clearer implementation brief with better scope, constraints, and completion criteria.

The AI coding mindset

The next step is learning how to think about the work in the right mode. Focused changes stay scoped to one area and can include many related tweaks, while bigger efforts need more planning, clearer scope, and better sequencing before implementation starts.

1

Bigger work

For bigger work: start with planning mode

When you are starting a project, redesigning a system, or dealing with open questions, planning mode helps you decide before code starts moving.

  1. 1 Turn on planning mode before you send your first prompt.
  2. 2 State the business or product goal first.
  3. 3 Ask the AI to think through design, content, data, technical changes, dependencies, testing, and rollout.
  4. 4 Ask the AI to call out missing decisions and tradeoffs.
  5. 5 Review the plan, approve the approach, and then move into implementation.
  6. 6 Once complete, follow-up changes work best in a new focused chat.

Planning reminders

  • Start a project in plan mode before asking for code.
  • Use planning mode when requirements are unclear or several systems are involved.
  • Good planning creates cleaner implementation chats later.
2

Focused changes

For focused changes: keep the chat scoped

Keep the chat scoped to one page, feature, or product area. You can make many related tweaks in one session — just avoid mixing unrelated areas in the same chat.

  1. 1 Describe the change in plain language and name the page, file, or workflow.
  2. 2 Add important constraints such as tone, design, or “do not break the current layout.”
  3. 3 Let the AI make the change and explain what it updated.
  4. 4 For a batch of related tweaks, commit when the area feels done.
  5. 5 For a standalone change, commit immediately before moving to something new.
  6. 6 When you are happy with the result, sync to GitHub to save and share your work.

Tip: Once you commit or sync, start a new chat for your next change.

Why focused, scoped chats help

  • Scope the chat to one area, not one change — related tweaks can share a session.
  • Fresh chats reduce confusion and make mistakes easier to spot.
  • Ask for verification only when the change affects builds, deploys, or public pages.

Prompt formula

A simple prompt formula

Before you write your prompt, choose the right mode. Then use this formula to give the AI what it needs. You do not need perfect wording, just enough context for the AI to understand the outcome and the finish line.

Choose your mode first

Choose your mode first. Use plan mode for bigger work with open questions. Use a focused chat for scoped changes in one area.

The CORD formula

Context + outcome + restrictions + done criteria

1. Plan mode

We are launching a new project management tool for small teams and need a marketing landing page to drive free trial signups. The page will live on our existing website. Keep the design modern and clean, consistent with our current brand, and do not change any other pages. Start in plan mode and create a PRD before making any changes. Before you begin, ask me any questions that would improve the plan. When done, give me a plan I can review and list any decisions you need me to make before work begins.

2. Focused chat

Update the contact page so the form sends an email confirmation on submit. The form is in the contact section of our marketing website. Keep the existing layout and button style, and do not change any other pages. Make the change now. When done, act as a user and validate the change works as expected. Keep updating until you are confident it will function correctly, then let me know what you did.

  • C Context: the page, file, workflow, or product area involved
  • O Outcome: what you want changed or built
  • R Restrictions: tone, layout, guardrails, or things not to break
  • D Done criteria: how you will know the work is complete (e.g., tests pass, page looks right, summarize what changed)

Copy-ready prompts

Copy-ready prompts

New project or bigger change

Enable plan mode first, then use this prompt to kick off planning before any implementation begins.

We are starting a new project. Review the current design and create a Product Requirements Document (PRD) before making changes.

Include goals, audience, key screens or user flows, design considerations, technical changes, dependencies, testing, rollout steps, and any missing decisions I should approve first.

Small app or website change

Use this for focused changes scoped to one page or product area.

Update this page so it does X. Keep the existing design approach, make only the necessary changes, and when you finish summarize what changed and flag anything that might affect other parts of the site.

If anything is unclear, ask only the questions needed to complete this one task.

Commit locally

Use this when you want to save your work but are not ready to push to production.

Commit all changes locally with a clear commit message describing what was done. Do not sync or push to GitHub. This work is not ready for production yet.

Commit and confirm

Use this after a change is ready to save your work and verify nothing broke.

Commit your changes, sync them to GitHub, and track GitHub Actions building my website to ensure there are no errors. When complete, provide me with the URL.

If anything fails, explain the error clearly and fix it before reporting back.

Final check

Before you hit send

A good AI coding prompt does not need to be long. It just needs the right starting mode, enough context, and a clear finish line.

  • Did I choose the right mode: plan mode or focused chat?
  • Did I include the context: the page, file, or workflow involved?
  • Did I describe the outcome: what I want changed or built?
  • Did I include any restrictions: tone, layout, or things not to break?
  • For small work, did I keep this to one focused task?
  • Do I want a commit, GitHub sync, or build verification at the end?