Study Notes — AI-Assisted Development

Vibe Coding &
Swarm Prototyping

The definitive guide to AI-assisted development methodologies — from conversational code generation (Vibe Coding) to multi-agent parallel prototyping (Swarm Prototyping). Understand when to use each, their strengths, risks, and how they're reshaping software engineering.

Updated: April 2026
Version: 1.0
Category: Foundation
Reading Time: ~15 min
Author: Michaël Bettan
01

What is Vibe Coding?

Vibe Coding is an AI-assisted development approach where the developer describes what they want in natural language, and an AI coding assistant generates the implementation. Coined by Andrej Karpathy (co-founder of OpenAI) in early 2025, the term captures a paradigm shift: instead of writing code character by character, you converse with an AI to produce working software.

The developer's role shifts from typing code to describing intent, reviewing output, and iterating through dialogue. You "vibe" with the AI — expressing what you want, accepting suggestions, nudging direction, and occasionally correcting course.

10×
Faster Prototyping
70%
Less Boilerplate
2025
Term Coined
∞
Use Cases

Core Principles of Vibe Coding

Key Insight

Vibe coding doesn't mean "no coding knowledge needed." The best vibe coders are experienced developers who leverage AI to amplify their expertise — they know what to ask for, how to evaluate the output, and when to intervene manually.

02

The Vibe Coding Spectrum

Vibe coding isn't binary — it exists on a spectrum from fully AI-driven to human-led with AI assistance. Understanding where you sit on this spectrum determines your workflow, risks, and tooling choices.

Full Autopilot

AI generates everything. Developer barely reads the code. Maximum speed, maximum risk.

  • Throwaway prototypes
  • Hackathon demos
  • Personal scripts
  • Risk: Very high

Collaborative Mode

Developer and AI co-create. Human reviews every block, steers architecture.

  • Feature development
  • Internal tools
  • MVPs for startups
  • Risk: Moderate

Human-Led + AI Assist

Developer writes core logic, AI handles boilerplate, tests, and docs.

  • Production systems
  • Security-critical code
  • Regulated industries
  • Risk: Low

Karpathy's Own Warning

Even Andrej Karpathy cautioned that vibe coding "is not for production systems" in its purest form. The more you move toward autopilot, the more you accumulate technical debt and hidden bugs you don't understand.

STEP 1

Describe

Write a natural language prompt describing what you want to build

STEP 2

Generate

AI produces code, architecture, or a full working component

STEP 3

Review

Developer reads, tests, and evaluates the generated output

STEP 4

Refine

Provide follow-up prompts to fix issues, add features, or refactor

STEP 5

Ship

Integrate into project, add tests, document, and deploy

03

What is Swarm Prototyping?

Swarm Prototyping takes vibe coding to the next level by deploying multiple AI agents simultaneously to build different variants, components, or approaches in parallel. Instead of one human conversing with one AI, you orchestrate a swarm of agents — each tackling a different aspect of the problem — and then select, merge, or compose the best results.

Think of it as A/B testing during development: spawn 5 agents to build 5 different implementations of the same feature, evaluate them all, and ship the winner. Or assign each agent a different service in a microservices architecture and build the entire backend simultaneously.

Key Characteristics of Swarm Prototyping

When to Use Swarm Prototyping

  • Exploring multiple architectural approaches
  • Building proof-of-concepts under time pressure
  • Hackathons and innovation sprints
  • Generating UI/UX design variants
  • Testing different algorithm implementations
  • Building microservices in parallel

When NOT to Use It

  • Tightly coupled systems requiring sequential design
  • Regulated/compliance-critical software
  • When integration costs exceed generation savings
  • Projects requiring deep domain expertise the AI lacks
  • When you can't evaluate quality of outputs
  • Production systems without human code review
04

Multi-Agent Architecture

Swarm prototyping relies on multi-agent systems (MAS) — architectures where independent AI agents collaborate, compete, or work in parallel. The design of the agent swarm determines the quality, speed, and reliability of the output.

Swarm Prototyping Architecture

Human
Orchestrator
→

Coordinator
Agent
→

Task
Decomposition
→

Agent Swarm
N workers
→

Merge &
Evaluate
Orchestrator Agent
The central coordinator that decomposes tasks, assigns work to specialized agents, and merges results. Can be human or AI.
Worker Agents
Specialized agents that execute specific tasks — e.g., one writes frontend, another writes backend, another writes tests.
Critic / Evaluator Agent
Reviews output from worker agents, scores quality, detects bugs, and recommends the best variant to proceed with.
Merger Agent
Combines outputs from multiple agents into a cohesive codebase, resolving conflicts and ensuring integration.

Swarm Patterns

05

Vibe Coding vs Swarm Prototyping

Vibe Coding

  • One human + one AI agent
  • Conversational, iterative
  • Sequential development
  • Human maintains full context
  • Low overhead, fast start
  • Best for focused tasks
  • Output: single implementation
vs

Swarm Prototyping

  • One human + many AI agents
  • Parallel, orchestrated
  • Concurrent development
  • Context split across agents
  • Higher overhead, needs setup
  • Best for exploration & scale
  • Output: multiple variants
Dimension Vibe Coding Swarm Prototyping
Agents Single AI assistant Multiple coordinated agents
Speed Fast for single tasks Faster for complex, multi-part projects
Exploration One path at a time Many paths simultaneously
Complexity Low setup, easy to start Requires orchestration infrastructure
Cost Low (one LLM session) Higher (N concurrent agents)
Integration Code is naturally coherent Requires merge/integration step
Best For Features, bug fixes, scripts Prototypes, architecture exploration, hackathons

They're Complementary, Not Competing

In practice, most teams use both: Vibe Coding for day-to-day development and Swarm Prototyping for exploration phases, design sprints, or when faced with ambiguous requirements where multiple approaches need to be evaluated quickly.

06

Tools & Ecosystem

Vibe Coding Tools

  • Cursor: AI-native IDE with inline chat, auto-complete, and multi-file context
  • GitHub Copilot: Inline code suggestions integrated into VS Code, JetBrains, Neovim
  • Claude Code (Anthropic): Agentic command-line coding assistant with tool use
  • Windsurf (Codeium): AI IDE with "Cascade" flows for multi-step coding tasks
  • Gemini Code Assist: Google's AI coding assistant with Gemini models
  • Aider: Open-source CLI pair programmer for git-tracked projects
  • Replit Agent: Full-stack AI agent that builds and deploys apps from prompts

Swarm / Multi-Agent Frameworks

  • OpenAI Swarm: Lightweight framework for multi-agent orchestration and handoffs
  • CrewAI: Framework for orchestrating role-based AI agents with structured workflows
  • AutoGen (Microsoft): Multi-agent conversations with customizable agent roles
  • LangGraph: Graph-based agent orchestration with state management
  • MetaGPT: Multi-agent framework simulating a software company (PM, architect, engineer)
  • Vertex AI Agent Builder: Google Cloud's managed platform for building agent systems
  • Amazon Bedrock Agents: AWS managed multi-step agent orchestration
Pseudo-code 
# Swarm Prototyping — Tournament Pattern
from swarm import Orchestrator, Agent, Evaluator

task = "Build a REST API for user authentication with JWT"

# Spawn 3 agents with different approaches
agents = [
    Agent(name="flask-agent",  style="Flask + SQLAlchemy"),
    Agent(name="fastapi-agent", style="FastAPI + Pydantic"),
    Agent(name="express-agent", style="Express.js + Passport"),
]

# All agents work in parallel
results = Orchestrator.fan_out(task, agents)

# Evaluator agent scores each implementation
best = Evaluator.rank(results, criteria=[
    "security", "code_quality", "test_coverage"
])

print(f"Winner: {best.name} — Score: {best.score}")
07

Risks, Limitations & Best Practices

Risks & Anti-Patterns

  • Cargo Cult Coding: Accepting AI output without understanding it — code works but you can't debug or extend it
  • Hidden Technical Debt: AI generates "good enough" code that accumulates complexity silently
  • Security Blind Spots: AI may introduce vulnerabilities (SQL injection, XSS, hardcoded secrets)
  • Hallucinated APIs: AI invents function calls or library methods that don't exist
  • Context Window Overflow: Large projects exceed AI context limits, causing incoherent outputs
  • Skill Atrophy: Over-reliance on AI may erode fundamental programming skills over time

Best Practices

  • Always Review: Read every line of generated code before committing — treat AI output as a draft
  • Test First: Write or generate tests before/alongside code — AI excels at test generation
  • Constrain Context: Provide clear, scoped prompts with architecture constraints
  • Version Everything: Git commit frequently — easy to rollback AI-generated changes
  • Security Scanning: Run SAST/DAST tools on all AI-generated code
  • Document Intent: Capture why you asked the AI for something, not just what it produced

The "It Works" Trap

The most dangerous outcome of vibe coding is code that appears to work perfectly but contains subtle logical errors, edge case failures, or security vulnerabilities that only manifest in production. Always ask: "Does it work, or does it seem to work?"

Production Readiness Checklist

08

Key Takeaways & Self-Assessment

Key Takeaways

Self-Assessment Questions

Q1. Who coined the term "Vibe Coding" and in what year?

Andrej Karpathy, co-founder of OpenAI, coined the term in early 2025 to describe the practice of conversational AI-assisted code generation.

Q2. What is the key difference between Vibe Coding and Swarm Prototyping?

Vibe Coding involves one human working with one AI agent conversationally. Swarm Prototyping orchestrates multiple AI agents working in parallel to explore different solutions or build different components simultaneously.

Q3. Name three swarm patterns and describe when you'd use each.

1) Fan-Out/Fan-In: Split a large task among agents, merge results — for decomposable projects. 2) Tournament: Agents compete on the same task, evaluator picks the best — for algorithm selection. 3) Specialization Grid: Each agent owns a domain (UI, API, DB) — for microservice architectures.

Q4. What is the "It Works" trap, and how do you mitigate it?

The "It Works" trap is when AI-generated code appears functional but contains hidden bugs, security vulnerabilities, or edge case failures. Mitigate by always reading every line, writing comprehensive tests, running security scans, and verifying all dependencies actually exist.

Q5. On the Vibe Coding spectrum, what are the three modes and their appropriate use cases?

1) Full Autopilot: Throwaway prototypes, hackathons, personal scripts (high risk). 2) Collaborative Mode: Feature development, MVPs, internal tools (moderate risk). 3) Human-Led + AI Assist: Production systems, security-critical code, regulated industries (low risk).
Study Progress — Vibe Coding & Swarm Prototyping 100%