Right. So here is the thing. The single biggest mistake I see new vibe coders make is treating Cursor or Antigravity like Google. They type a one-line prompt. The AI guesses. They get a half-broken app. They spend the next two hours and 6,000 credits arguing with the AI to fix what it never understood in the first place. Trust me on this one. I have watched it happen in every single cohort.
The fix is templates. Specifically, templates that flip the loop. Instead of you guessing what to tell the AI, the AI asks you the right questions first. Then builds.
These are the 8 patterns I teach in my Digital Scholar Vibe Coding cohort. They are the same prompts I use on paid client work as a vibe coding expert in India. Every one of them has shipped real production apps. Use them as starting points. Adapt them. Save your filled-in versions for next time.
The framework: ask first, build second
Every template in this article shares one DNA. The first instruction tells the AI to ask you a list of clarifying questions one at a time and wait for your answer before moving on. Then it produces a plan. You review the plan. Only after approval does the AI write code.
Three rules:
- Specific answers, vague answers. Vague answers produce vague builds. If you do not know an answer, tell the AI and ask for suggestions instead of guessing.
- Always review the plan before approving. This is the last checkpoint before credits start burning on code generation.
- Save your answered versions. The next time you build a similar project, start from what worked, not from blank.
The 8 templates
Full e-commerce web application
Forces the AI to ask 10 clarifying questions before writing any code. Customer category, the real business problem, target buyer, brand tone, features, variants, checkout style, payments, blog needs, accounts. Every answer compounds into a sharper plan and fewer credits burned later.
Landing page for a service business
Asks 8 questions about your service, ideal client, the single action you want a visitor to take, your benefits, testimonials, brand colours, contact preferences, and ad-traffic context. Pair it with Fast mode for a build under 15 minutes.
Inquiry-based B2B catalogue site
Captures the difference: spec sheets, compliance certs, datasheets, MOQ. Inquiry form replaces checkout. Routes lead to email, WhatsApp, or your CRM.
Add a feature to an existing build
Forces the AI to summarise its understanding and list every file it plans to touch before writing code. This single step saves you most of the 'oh no, the cart broke' moments.
Fix something that is broken
Debugging template. Asks for the page, expected behaviour, actual behaviour, repro steps, frequency, error messages. The AI then proposes a root-cause fix before changing code.
Redesign an existing page
Asks what to keep, what to change, the mood, mobile vs desktop priority, and reference screenshots. Outputs a layout plan you approve before any pixels move.
Add admin dashboard features
Covers user roles, activity logs, confirmation gates on destructive actions, reports the client actually wants. Saves dozens of post-launch support requests.
Integrate a third-party tool
Asks for the tool, purpose, API keys, trigger event, data flow, fallback behaviour. Then plans the files and env vars needed before touching code.
How to use them daily
- Plan the project on paper before opening the IDE. 5 minutes here saves 30 minutes of credits later.
- Pick the right template. New project: 1, 2, or 3. Existing project: 4, 5, 6, 7, or 8.
- Open Antigravity (or Cursor / Claude Code) in Planning mode. Always Planning mode for new projects.
- Pick the right model. Gemini 3 Pro for UI and design. Claude Sonnet 4.6 for backend and debugging.
- Paste the template, answer the questions, review the plan, approve. Then the AI builds.
- Use screenshots to communicate visual issues. A screenshot plus 'fix this' beats a 200-word description every time.
- Switch models when one credit pool is exhausted. The other pool keeps you working without waiting 4 to 5 days for a refresh.
What separates a vibe coding expert in India from someone just typing prompts
Anybody can paste a template. The expert knows when to break the template. When the AI's plan is good enough, when to push back, when to call a section done, when to scrap the whole thing and start fresh. That judgement comes from shipping. Real shipping, to real users, with real consequences when something breaks.
If you are early in your vibe coding journey, do not skip the boring projects. Build the todo app. Build the invoice generator. Build the bookmark manager. Each one teaches you a piece of the judgement that no template can. That is the actual moat.
FAQ
Why do these templates work better than freestyle prompts?+
Because they front-load the detail. The biggest credit-burn pattern I see is people writing a vague prompt, the AI making 10 wrong assumptions, then 30 minutes of credit-eating back-and-forth. These templates flip the loop. The AI asks the questions first, you answer once, then it builds. Same output, 1/3 the credits.
Can I use these with Cursor and Claude Code, not just Antigravity?+
Yes. The templates work in any AI coding tool. The 'ask me questions first' pattern is universal. Cursor, Claude Code, Antigravity, even Lovable and v0 if you tweak the wording. I run the same patterns across all of them.
Do you give the full prompts to your cohort?+
Yes. The Digital Scholar Vibe Coding cohort gets the complete prompt text plus my answered versions from real client builds, plus voice-overs explaining why each line matters. This blog gives you the framework. The cohort gives you the artefacts.
Do I need to be technical to use these templates?+
No, but you need to be able to read what the AI gives you and recognise when something looks off. That is the 'vibe' part of vibe coding. It is judgement, not syntax. Anyone smart can learn it. We cover this in the cohort and on 1:1 fractional engagements.
How is Eugene Samuel different from other vibe coding experts in India?+
I ship for paying clients every week. I am not just a course seller. The prompt templates in this article are the same ones I use on real fractional engagements where the deliverable is a working production app, not a demo. Everything I teach has been proven on a paid invoice.
Want the full templates with my answered versions?
Join the Digital Scholar Vibe Coding cohort or book a 30-min fractional discovery call. Either way you get the artefacts plus the judgement.
See Eugene's Vibe Coding Programs