Latest / AI Builds It: Easy Coding Tools / Your First Prompt for an AI Coding Agent: What to Say and What to Leave Out

Transcript

Your First Prompt for an AI Coding Agent: What to Say and What to Leave Out · AI Builds It: Easy Coding Tools

• 0:00
  SPEAKER_00: Your first prompt for an AI coding agent.
• 0:03
  SPEAKER_00: What to say and what to leave out.
• 0:05
  SPEAKER_00: A coding agent can turn a plain language request into a working website, but it cannot read your mind.
• 0:12
  SPEAKER_00: Make me a nice website leaves big choices open.
• 0:16
  SPEAKER_00: Who is it for?
• 0:17
  SPEAKER_00: What should it do?
• 0:18
  SPEAKER_00: What does nice look like?
• 0:20
  SPEAKER_00: A useful prompt answers those questions in simple terms.
• 0:24
  SPEAKER_00: It also says what the agent should not build and when it should stop.
• 0:28
  SPEAKER_00: That matters even more when revisions use credits or count toward a usage limit.
• 0:33
  SPEAKER_00: A clear first request gives the agent a better target.
• 0:37
  SPEAKER_00: A clear stopping point helps prevent it from adding features you never asked for.
• 0:42
  SPEAKER_00: This guide is based on the tools published docs and announcements checked on October 12, 2026.
• 0:49
  SPEAKER_00: The difference between a vague prompt and a useful one.
• 0:52
  SPEAKER_00: Vague.
• 0:53
  SPEAKER_00: Make me a professional website for a dog walker.
• 0:56
  SPEAKER_00: Make it look good and include everything a dog owner needs.
• 1:00
  SPEAKER_00: This sounds clear to a person, but the agent still has to guess what professional means, which pages to make, what the site should help visitors do, and what everything includes.
• 1:10
  SPEAKER_00: More useful.
• 1:11
  SPEAKER_00: Build a one-page website for a local dog walking service.
• 1:15
  SPEAKER_00: It is for busy dog owners who want to learn about services and send a booking inquiry.
• 1:20
  SPEAKER_00: Include a welcome section, three service cards, a service area section, a short FAQ, and a contact form.
• 1:27
  SPEAKER_00: The main button should take visitors to the form.
• 1:30
  SPEAKER_00: Use a warm, friendly look with a cream background, dark blue text, and green accents.
• 1:35
  SPEAKER_00: Make it easy to read on a phone.
• 1:37
  SPEAKER_00: Use fictional sample text and contact details.
• 1:40
  SPEAKER_00: The form should show a success message, but should not send or store anything.
• 1:45
  SPEAKER_00: Build only this one page.
• 1:47
  SPEAKER_00: Do not add login, payments, a database, or extra pages.
• 1:51
  SPEAKER_00: Check the page on phone and desktop sizes.
• 1:53
  SPEAKER_00: Stop when those parts work.
• 1:55
  SPEAKER_00: Then tell me what you built.
• 1:57
  SPEAKER_00: This request gives the agent four things it needs the audience, the features, the appearance, and the finish line.
• 2:04
  SPEAKER_00: Replit's own prompting guide also recommends clear requirements, examples, and smaller steps instead of broad requests.
• 2:11
  SPEAKER_00: Replit's Guide to Efficient Prompting.
• 2:13
  SPEAKER_00: What to put in your first prompt.
• 2:15
  SPEAKER_00: Name the audience and the goal.
• 2:17
  SPEAKER_00: Start with who the website is for and what you want visitors to do.
• 2:20
  SPEAKER_00: For example, this page is for local dog owners.
• 2:24
  SPEAKER_00: The main goal is to get booking inquiries.
• 2:26
  SPEAKER_00: Visitors should be able to compare services and find the service area.
• 2:30
  SPEAKER_00: This helps the agent make useful choices.
• 2:32
  SPEAKER_00: A site for people ready to book should not bury the contact button beneath a long company story.
• 2:38
  SPEAKER_00: List the features that matter now.
• 2:40
  SPEAKER_00: Choose a short list of must-haves.
• 2:42
  SPEAKER_00: For a simple business website, that might be a welcome section, a list of services, a clear contact button, a short FAQ, a contact form.
• 2:51
  SPEAKER_00: Add a contact form is less useful than add a form with name, email, and message fields.
• 2:58
  SPEAKER_00: Show a success message when the visitor submits it.
• 3:02
  SPEAKER_00: Do not ask for everything a real business needs.
• 3:06
  SPEAKER_00: That phrase can grow into user accounts, payments, staff tools, and a database.
• 3:11
  SPEAKER_00: Ask for the smallest version that proves your idea.
• 3:15
  SPEAKER_00: Describe the look in plain words.
• 3:18
  SPEAKER_00: You do not need design jargon.
• 3:20
  SPEAKER_00: Give the agent a few clear choices.
• 3:23
  SPEAKER_00: Mood, calm, playful, warm, bold, colors, cream background, dark blue text, green accents.
• 3:30
  SPEAKER_00: Layout, one page, clear sections, large buttons.
• 3:34
  SPEAKER_00: Reading experience, easy to scan on a phone.
• 3:36
  SPEAKER_00: Make it modern is not a full design brief.
• 3:39
  SPEAKER_00: A few visual details give the agent a direction without telling it every pixel.
• 3:44
  SPEAKER_00: Set a finish line.
• 3:46
  SPEAKER_00: Tell the agent when the first version is done.
• 3:48
  SPEAKER_00: This is the part many beginners leave out.
• 3:51
  SPEAKER_00: For example, stop after building the one-page website, checking the form's demo response, and confirming the page fits phone and desktop screens.
• 4:00
  SPEAKER_00: Do not add other features.
• 4:01
  SPEAKER_00: Give me a short list of what works and what still needs a real service behind it.
• 4:06
  SPEAKER_00: A finish line makes revision easier too.
• 4:09
  SPEAKER_00: You can judge the result against the request instead of asking, is this done yet?
• 4:14
  SPEAKER_00: What to leave out?
• 4:15
  SPEAKER_00: Your first prompt does not need a long technical plan.
• 4:18
  SPEAKER_00: Usually, leave out.
• 4:20
  SPEAKER_00: Coding terms you don't understand.
• 4:22
  SPEAKER_00: Let the agent choose the simple technical setup unless you have a real reason to choose it.
• 4:27
  SPEAKER_00: A long wish list.
• 4:28
  SPEAKER_00: Build the main page first.
• 4:30
  SPEAKER_00: Add one feature at a time.
• 4:32
  SPEAKER_00: Vague demands.
• 4:33
  SPEAKER_00: Making amazing gives the agent no clear test.
• 4:36
  SPEAKER_00: Secrets or private information.
• 4:38
  SPEAKER_00: Do not paste passwords, API keys, payment details, customer records, or private business data into a prompt.
• 4:46
  SPEAKER_00: Use fictional examples.
• 4:47
  SPEAKER_00: If an app later needs a secret key, use the tool's secure secrets feature, not the chat.
• 4:53
  SPEAKER_00: Lovable's Guide to Project Secrets.
• 4:55
  SPEAKER_00: A request to make a simple website production ready.
• 4:58
  SPEAKER_00: That can mean many things, from security checks to live payments.
• 5:03
  SPEAKER_00: Name the exact task you want.
• 5:05
  SPEAKER_00: For a first website, it is smart to say what should stay out.
• 5:08
  SPEAKER_00: No login, no payment flow, no database, and no extra pages.
• 5:13
  SPEAKER_00: Those are not forever rules.
• 5:15
  SPEAKER_00: They are ways to keep the first version small.
• 5:18
  SPEAKER_00: A reusable prompt for a simple website.
• 5:20
  SPEAKER_00: Copy this prompt and replace the bracketed parts.
• 5:23
  SPEAKER_00: Keep only the features you truly need.
• 5:26
  SPEAKER_00: The line, stop after this first version is not magic.
• 5:29
  SPEAKER_00: But paired with a clear list of requirements, it gives the agent a practical boundary.
• 5:34
  SPEAKER_00: How follow-up instructions work in each tool.
• 5:37
  SPEAKER_00: You do not have to get every detail right in the first message.
• 5:40
  SPEAKER_00: These tools let you steer the work afterward.
• 5:43
  SPEAKER_00: The smart move is to make each follow-up about one visible change.
• 5:47
  SPEAKER_00: Replit agent can build from a natural language request, then take follow-up instructions in the project's conversation.
• 5:55
  SPEAKER_00: For a larger task, use plan mode to review the proposed approach before asking it to build.
• 6:08
  SPEAKER_00: Replit's prompting guide, replit checkpoints, and version control.
• 6:12
  SPEAKER_00: Replit's August 18th, 2026 free mode announcement added a way to explore and build within usage limits.
• 6:20
  SPEAKER_00: That is useful for trying an idea, but it does not mean every revision is unlimited or guaranteed to be free.
• 6:27
  SPEAKER_00: Current Replit docs list free, power, and max modes.
• 6:31
  SPEAKER_00: Power is aimed at cost-conscious work, while Max uses more tapable models for harder tasks.
• 6:37
  SPEAKER_00: Check the mode and usage display before a long run.
• 6:40
  SPEAKER_00: Replit's free mode announcement.
• 6:42
  SPEAKER_00: Current Replit agent modes.
• 6:45
  SPEAKER_00: Good Replit follow-up.
• 6:46
  SPEAKER_00: Change the main button from blue to green.
• 6:49
  SPEAKER_00: Keep its wording, size, and position the same.
• 6:52
  SPEAKER_00: Do not change anything else.
• 6:54
  SPEAKER_00: Lovable Lovable lets you send follow-ups in the project chat and steer the work while it is underway.
• 7:01
  SPEAKER_00: Use chat mode to discuss an idea without changing code, plan mode to review a proposed change before implementation, and build mode to make changes.
• 7:10
  SPEAKER_00: Lovable also provides version history for restoring earlier work.
• 7:14
  SPEAKER_00: Lovable project chat, plan mode, build mode, version history, a later update added drafts, which let you explore a separate version without changing the live project.
• 7:25
  SPEAKER_00: That can be helpful when you want to compare two design ideas before choosing one.
• 7:30
  SPEAKER_00: Lovable Drafts.
• 7:31
  SPEAKER_00: On June 13th, 2026, Lovable announced one credit balance for building and running apps, along with a clearer billing page.
• 7:40
  SPEAKER_00: The company said the plan price, credit cost, and rate of credit use were not changing.
• 7:45
  SPEAKER_00: So the update made spending easier to track.
• 7:48
  SPEAKER_00: It did not promise cheaper revisions.
• 7:50
  SPEAKER_00: Lovable's billing update.
• 7:52
  SPEAKER_00: Good lovable follow-up.
• 7:54
  SPEAKER_00: In the welcome section only, shorten the paragraph to two sentences.
• 7:58
  SPEAKER_00: Keep the meaning and design.
• 8:00
  SPEAKER_00: Do not change the other sections.
• 8:02
  SPEAKER_00: Clawed code.
• 8:03
  SPEAKER_00: Clawed code works with the project's files, so it is a more hands-on option than a simple build chat.
• 8:09
  SPEAKER_00: It is available in several settings, including the terminal, IDEs, desktop, and browser.
• 8:15
  SPEAKER_00: For a change that may affect several files, use plan mode first.
• 8:19
  SPEAKER_00: Review its plan, then ask it to make the change.
• 8:22
  SPEAKER_00: You can give new instructions in the session, and Claude Code's checkpoints can help you rewind edits.
• 8:27
  SPEAKER_00: Clawed code overview, Claude Code permission modes, Claude Code Checkpoints.
• 8:32
  SPEAKER_00: The cost of a revision depends on how you use Claude Code.
• 8:36
  SPEAKER_00: A subscription has usage limits.
• 8:38
  SPEAKER_00: API access is billed by model use.
• 8:41
  SPEAKER_00: On May 6, 2026, Anthropic said it had doubled the 5-hour Claude Code rate limits for several paid plans.
• 8:49
  SPEAKER_00: In August, Anthropic also announced that Sonnet 5's API pricing would remain at$2 per million input tokens and$10 per million output tokens.
• 9:01
  SPEAKER_00: Those are API rates, not a fixed price for every edit.
• 9:05
  SPEAKER_00: A small text change and a broad code-based task can use different amounts.
• 9:10
  SPEAKER_00: Anthropic's usage limit update.
• 9:17
  SPEAKER_00: Update only the mobile spacing in the welcome section.
• 9:20
  SPEAKER_00: Keep the desktop layout and all text unchanged.
• 9:24
  SPEAKER_00: Show me the changes when you finish.
• 9:26
  SPEAKER_00: Did summer 2026 change how revisions work?
• 9:30
  SPEAKER_00: A little, but not enough to change the basic advice.
• 9:33
  SPEAKER_00: Replit's free mode changed how people can start and explore.
• 9:37
  SPEAKER_00: Lovable's June billing update combined balances and improved cost visibility while keeping the stated credit rates the same.
• 9:45
  SPEAKER_00: Anthropic increased some clawed code usage limits, but that was not a new flat price for revisions.
• 9:51
  SPEAKER_00: The practical lesson is the same across all three tools.
• 9:54
  SPEAKER_00: Small, clear follow-ups are easier to judge than redo the whole thing requests.
• 10:00
  SPEAKER_00: Tell the agent what to change, where to change it, and what to leave alone.
• 10:04
  SPEAKER_00: Then check whether the change worked before giving it the next task.
• 10:08
  SPEAKER_00: Conclusion.
• 10:09
  SPEAKER_00: Give the agent a brief, not a blank check.
• 10:12
  SPEAKER_00: Your first prompt does not need to be clever or technical.
• 10:15
  SPEAKER_00: Tell the agent who the site is for, what it must do, what it should look like, and when to stop.
• 10:21
  SPEAKER_00: Leave out the long wish list, leave out private data.
• 10:24
  SPEAKER_00: Start with one useful page, then improve it one focused request at a time.
• 10:29
  SPEAKER_00: That is a much better path than asking for everything and hoping the agent guesses right.
• 10:34
  SPEAKER_00: All links to sources are available in the text version of this article.
• 10:38
  SPEAKER_00: You can find the full article at easycoding.tools slash blog.
• 10:43
  SPEAKER_00: Thanks for listening.
• 10:44
  SPEAKER_00: For more practical AI coding guides, tool comparisons, and resources for builders, visit easycoding.tools.
• 10:51
  SPEAKER_00: And if it's easier to remember, just go to aibuilds.it.
• 10:56
  SPEAKER_00: It redirects to easycoding.tools.