Skip to content
BSWEN
Blog

Vibe Coding with AI: How I Completed a 3-Week Project in 2 Days

Cowrie
Cowrie
Dev @ Bswen

Problem

My lead designer handed me a project estimated at 2.5-3 weeks of work. As a lead programmer at a small game studio, I normally would:

  1. Spend days analyzing requirements
  2. Build data models
  3. Implement features one by one
  4. Debug and fix issues
  5. Coordinate with my 3-4 junior developers

But I tried something different. I used AI to write the code for me.

Result: 2 days. The entire project, done.

This is what people call “vibe coding”—and it changed how I think about programming.

Environment

  • Game development studio
  • Python backend, game-specific logic
  • AI tools: GPT, Gemini, Kimi, DeepSeek, Qwen
  • Project scope: New game feature with multiple components

What Is Vibe Coding?

“Vibe coding” means you guide AI to generate code through natural language descriptions. Instead of writing every line yourself, you:

  1. Understand what needs to be built
  2. Describe it to AI in clear terms
  3. Review the generated code
  4. Request fixes for bugs
  5. Deploy

Your role shifts from code writer to code director.

What Happened?

Day 1: Full Implementation

I received requirements from our game designers. The document was messy—game designers aren’t known for technical precision.

My old approach:

text title=“Traditional Workflow”

  1. Read requirements (hours)
  2. Clarify ambiguities (more hours)
  3. Design data models (half day)
  4. Write code (days)
  5. Debug issues (more days)
  6. Review with team (yet more time)

My new approach:

text title=“Vibe Coding Workflow”

  1. Read requirements
  2. Translate into AI prompts (specify architecture, constraints)
  3. AI generates code (seconds)
  4. Review output, find bugs
  5. Report bugs to AI, receive fixes
  6. Done

The AI could understand 70-80% of requirements even from raw designer documents. I only needed to clarify the edge cases.

Day 2: Requirement Changes

Game designers always change their minds. Normally this means rework.

With AI, I just described the changes. The AI updated the code in minutes.

The Only Manual Code Change

The only thing I wrote by hand? Renaming enum values.

python title=“AI Generated” class PlayerStatus(Enum): AUTHENTICATED = “AUTHENTICATED” IN_GAME = “IN_GAME”

python title=“My Manual Edit” class PlayerStatus(Enum): LOGIN = “login” # Easier to read PLAYING = “playing” # Less “professional” but clearer

The AI’s English was “too professional”—harder for our team to scan quickly.

How Vibe Coding Works

The Prompt Structure

When I prompt AI for code, I follow this pattern:

text title=“Prompt Template” “Create a [component name] for [purpose].

Include:

  • [feature 1]
  • [feature 2]
  • [feature 3]

Use [architecture pattern] we discussed. Add [specific constraints].”

Example: Player Management

text title=“My Prompt to AI” “Create a PlayerManager class that handles player lifecycle in a multiplayer game.

Include:

  • Player registration with duplicate checking
  • Session management with timeout
  • Statistics tracking
  • Thread-safe operations

Use the repository pattern we discussed. Handle up to 10,000 concurrent players.”

The AI generated the entire class in seconds. I reviewed it, tested it, found one edge case with reconnection, asked AI to fix it, and moved on.

What I Actually Do Now

text title=“My Real Work” ┌─────────────────────────────────────────────┐ │ What AI Does (90%) │ ├─────────────────────────────────────────────┤ │ • Write all boilerplate code │ │ • Generate CRUD operations │ │ • Implement standard patterns │ │ • Fix bugs when I report them │ │ • Update code for requirement changes │ └─────────────────────────────────────────────┘ ↓ ┌─────────────────────────────────────────────┐ │ What I Do (10%) │ ├─────────────────────────────────────────────┤ │ • Understand business requirements │ │ • Design architecture decisions │ │ • Translate requirements to AI prompts │ │ • Review AI output for correctness │ │ • Identify edge cases AI misses │ │ • Make judgment calls on trade-offs │ └─────────────────────────────────────────────┘

Why This Matters

The Emotional Value

Here’s something I didn’t expect: AI provides better emotional value than my subordinates.

When I report a bug to AI:

  • AI acknowledges the error
  • AI fixes it immediately
  • AI apologizes if needed
  • No drama, no excuses, no delays

When I report a bug to a junior developer:

  • Defensive explanations
  • “It works on my machine”
  • Delayed fixes
  • Back-and-forth debugging

The Productivity Math

text title=“Team Comparison” Traditional team (me + 3 juniors):

  • 3 weeks for project
  • 4 salaries
  • Communication overhead
  • Code review bottleneck

AI-augmented (me + AI):

  • 2 days for same project
  • 1 salary + API costs
  • No communication overhead
  • Instant iterations

One person with AI can now do the work of three developers—faster and with fewer errors.

Common Mistakes

Mistake #1: Using AI Only as Search Engine

Before mid-2025, I used AI to:

  • Find documentation
  • Get code snippets
  • Understand error messages

I was underestimating AI. The real power is full code generation, not just assistance.

Mistake #2: Not Specifying Architecture

If you just say “write a login system,” AI will generate something generic. You need to specify:

  • What architecture pattern to use
  • What constraints to follow
  • What existing code to integrate with

Mistake #3: Blind Trust

AI makes mistakes. You must:

  • Review all generated code
  • Test edge cases
  • Validate against requirements
  • Catch security issues

Summary

In this post, I showed how vibe coding with AI delivered 10x productivity improvement. The key point is that my role shifted from writing code to directing AI-generated code.

The workflow:

  1. Understand requirements deeply
  2. Translate to clear AI prompts
  3. Review and validate output
  4. Iterate on fixes

What changed for me:

I used to be a “big soldier”—doing the same coding work as my team, just faster. Now I’m a translator between business needs and AI capability.

The future of programming isn’t about who writes the best code. It’s about who can best direct AI to write code.

Final Words + More Resources

My intention with this article was to help others share my knowledge and experience. If you want to contact me, you can contact by email: Email me

Here are also the most important links from this article along with some further resources that will help you in this scope:

Oh, and if you found these resources useful, don’t forget to support me by starring the repo on GitHub!

Comments

Tags: