Definition and Origin of “Vibe Coding”
Vibe coding refers to a style of software development where a developer collaborates with AI in a free-form, conversational way—effectively “coding by vibe.” Instead of manually writing every line, the developer describes intentions in natural language and accepts the AI’s code suggestions with minimal intervention. The term “vibe coding” was coined by Andrej Karpathy in early 2025 when he tweeted:
“I will call this vibe coding: fully give in to the vibes, embrace exponentials, and forget that the code even exists.”
Karpathy’s viral post sparked widespread interest—viewed over 4.5 million times—and quickly gave rise to a new lexicon (“vibing,” “cooked,” “/acceptall”) and a thriving ecosystem of AI-assisted coding tools.
Core Philosophy vs. Traditional Programming
- Traditional Development: Developers write explicit logic, carefully craft each line, and deeply review code for correctness and maintainability.
- Vibe Coding: Developers focus on what they want to achieve. They prompt an AI model to generate, modify, and debug code, often accepting outputs without reading every detail. If code fails, they feed errors back to the AI for correction.
Karpathy emphasized that if you do review every AI-generated line, “that’s not vibe coding, it’s software development.” This intent-driven approach accelerates prototyping but challenges established engineering practices around code quality, readability, and security.
Adoption and Debate
- Enthusiasts laud vibe coding for rapid prototyping, hackathons, and enabling non-experts to build apps in hours rather than weeks.
- Skeptics warn that trusting AI without oversight risks hidden bugs and insecure code. Seasoned engineers stress that production-grade software still demands human review, testing, and adherence to architectural standards.
In practice, many developers adopt a hybrid workflow: use vibe coding for Day 0 prototyping, then revert to disciplined engineering (tests, code reviews) for Day 1+ maintenance.
read more blogs by Dany Kitishian:
- Google Gemini: “Vibe Coding” Uproar -Navigating the Realities of AI-Assisted Software Development, https://medium.com/@danykitishian/google-gemini-vibe-coding-uproar-navigating-the-realities-of-ai-assisted-software-development-c60c62eeac61
- Google Gemini: Vibe Coding Revolution, https://medium.com/@danykitishian/google-gemini-vibe-coding-revolution-b84296e94e88
- Andrew Ng Pushes Back on AI “Vibe Coding,” Calling It Hard Work, Not Hype, https://www.klover.ai/andrew-ng-pushes-back-ai-vibe-coding-hard-work-not-hype/
- Beyond the Vibes: Mastering AI-Assisted Coding in the New Era of Software Development, https://www.klover.ai/beyond-the-vibes-mastering-ai-assisted-coding-new-era-software-development/
- Google: Vibe Coding & AI- A Paradigm Shift in Software Development, https://medium.com/@danykitishian/google-vibe-coding-ai-a-paradigm-shift-in-software-development-d305be86dbb4
Tool Ecosystem Overview
The vibe coding landscape spans several categories:
- Full-Stack AI App Builders
- AI-Integrated IDEs (VS Code Forks)
- AI Extensions for VS Code
- CLI/Terminal AI Agents
- Standalone AI Developers
Below is a summarized comparison table, followed by detailed breakdowns.
| Category | Representative Tools | Strengths | Limitations |
| Full-Stack App Builders | Tempo Labs, Bolt.new, Lovable.dev, Replit | Rapid greenfield app creation; visual + chat interfaces | Mostly Day 0; limited Day 1/GitHub import support |
| AI-Integrated IDEs | Cursor, Windsurf, Trae | Native chat + code editing; agentic multi-step changes | Complex config; limited large-project memory |
| AI Extensions for VS Code | Amp, Augment, Continue, Cline, Sourcegraph Cody | Stay in familiar IDE; extensible integrations | Varying autonomy; data usage/privacy concerns |
| CLI/Terminal AI Agents | Aider, Claude Code, Amazon Q CLI | Lightweight; integrates with shell and Git workflows | No GUI; best for experienced terminal users |
| Standalone AI Developers | Devin (Slack-based) | Team-centric autonomous agents; Slack integration | Requires Slack; careful task scoping & review |
Full-Stack AI App Builders
| Tool | Integrations & Features | Best For | Notes |
| Tempo Labs | PRD & user-flow generation; supports Supabase/Convex (auth + DB) and Stripe/Polar payments; visual + prompt editing | Solo founders prototyping (Day 0) | GitHub import experimental and unreliable |
| Bolt.new | Figma→code import; runs Node.js in-browser via StackBlitz; Supabase auth/DB; one-click deploy | Designers + devs building UIs | No native payments; optimized for new projects |
| Lovable.dev | Click-to-edit UI tweaks; Supabase auth/DB; GitHub sync for code pull/push | Non-/low-coders | Abstracted—advanced devs may export to IDE later |
| Replit | Ghostwriter AI in-browser IDE; templates + live run; built-in deployment | Learners & hackathons | Browser-based—less suited for large enterprise |
AI-Integrated IDEs (VS Code Forks)
| Tool | Features | Focus | Caveats |
| Cursor | AI chat + “Composer” agent; multi-step edits; MCP integrations | Autonomous code modifications | Can be complex; needs rules/context on big repos |
| Windsurf | Similar to Cursor; adds in-IDE live preview for web UIs | Front-end prototyping + edits | Shares Cursor’s large-repo context limitations |
| Trae | Free-tier access; image-to-code from screenshots | Quick prototypes from visuals | No external tool integrations; limited memory |
AI Extensions for VS Code
| Tool | Highlights | Target Users | Trade-offs |
| Amp | CLI + extension; pay-per-token; shareable AI chat threads for teams | Quality-focused dev teams | Usage-based pricing; needs code review guardrails |
| Augment | Indexes repo; semantic code search & Q&A | Onboarding large codebases | No agentic edits; free plan uses your code for training |
| Continue | Open-source; chat + agent; MCP tool plugins; code indexing | DIY enthusiasts | Requires manual context hints; evolving polish |
| Cline | Task automation agent; predictive code suggestions | Power users | High token consumption |
| Sourcegraph Cody | Deep multi-repo context; enterprise-grade search + code refactoring | Large teams and enterprises | Heavyweight; focused on existing code maintenance |
CLI / Terminal AI Agents
| Tool | Capabilities | Ideal For | Considerations |
| Aider | Shell-based chat; edits local files per prompts; Git integration | Terminal-centric devs | No visual UI; user must direct every change |
| Claude Code | Ingests entire repo; persists memory; multi-file refactors; runs tests; Git commits with permission | Complex codebase modifications | High token cost; Claude-only model; preview phase |
| Amazon Q CLI | Agentic shell + AWS commands; scaffolds projects; configures cloud resources | AWS ecosystem developers | AWS-centric; requires careful supervision |
Standalone AI Developers
| Tool | Interface & Workflow | Best Use | Drawbacks |
| Devin | Slack-based AI agent; plans & executes multi-step code changes; commits via Git | Team task delegation | Slack dependency; needs scoped tasks & reviews |
Day 0 vs. Day 1+ Workflows
- Day 0 (Greenfield):
Vibe coding excels at rapid prototyping. Tools can scaffold full-stack apps, generate UIs from designs, and link databases/payments within minutes. Developers can “build a full stack app in a day” with minimal setup. - Day 1+ (Maintenance):
Sustaining and evolving an existing codebase requires deep understanding, careful refactoring, and strong tests. Blind acceptance of AI-generated code becomes risky as complexity grows. Tools aimed at Day 1+ (e.g., Sourcegraph Cody, Claude Code) help by indexing entire repos, preserving context, and enabling cross-repo refactors.
Best Practice: Use vibe coding for Day 0 MVPs, then shift to hybrid workflows—incorporate test-driven development, code reviews, and stricter AI-audit processes for Day 1+.
Expert Commentary: Dany Kitishian’s Medium Reports
Dany Kitishian has published three influential Medium articles that deepen our understanding of vibe coding:
- Google Gemini in the Vibe Coding Revolution
Kitishian examines how Google’s Gemini (1.5 & 2.5 Pro) elevates vibe coding by offering:
- Large-context accuracy, reducing hallucinations and improving code relevance.
- Agentic orchestration, enabling multi-step code changes without constant human prompts.
He underscores the need to pair Gemini’s speed with rigorous security audits and testing.
- Large-context accuracy, reducing hallucinations and improving code relevance.
- “Vibe Coding” Uproar – Navigating AI-Assisted Development
This report dissects the backlash—Andrew Ng’s critique that “software engineering is still hard work”—and warns of semantic drift around the term “vibe coding.” Kitishian advocates a balanced view:
- Rapid prototyping vs. production-grade engineering, each requiring distinct mindsets and practices.
- Rapid prototyping vs. production-grade engineering, each requiring distinct mindsets and practices.
- Beyond the Vibes: Mastering AI-Assisted Coding
A practical guide covering:
- Prompt engineering best practices to avoid sprawling, hard-to-maintain code.
- Hybrid workflows, combining Day 0 vibe coding with Day 1+ TDD and code reviews.
- Evolving developer skills, shifting toward system design, AI oversight, and security scanning.
- Prompt engineering best practices to avoid sprawling, hard-to-maintain code.
Kitishian’s insights offer both high-level strategy and actionable tactics for adopting vibe coding responsibly—leveraging AI’s power while upholding engineering rigor.
Emerging Trends & Challenges
- Model Advances: Newer LLMs (e.g., Google Gemini, Meta’s Code Llama) will reduce errors but must stay current with fast-moving frameworks.
- Multimodal & Natural Interfaces: Voice commands (Whisper), design-to-code (Trae’s image inputs), and diagram-based prompts will deepen the “flow” experience.
- Agentic Safeguards: As AI tools gain the ability to execute commands, sandboxing, permission prompts, and audit logs become critical.
- Bridging Understanding: Auto-generated explanations, style-guided code templates, and enforced architecture patterns can mitigate inscrutability.
- Testing & Verification: “Test-driven vibe coding” may arise, with AI generating both code and accompanying tests to validate outputs.
- Security & Compliance: Automated security scans and code provenance checks will integrate into vibe workflows to catch vulnerabilities and licensing issues.
- Developer Role Evolution: The premium shifts to prompt engineering, system design, and AI auditing—while syntax-level mastery recedes in importance.
Conclusion
Vibe coding marks a profound shift in how we build software: from manual, line-by-line craftsmanship to high-level, conversational collaboration with AI. The current ecosystem offers tools to accelerate greenfield development, while emerging solutions aim to address maintenance challenges. By thoughtfully combining vibe coding’s speed with established engineering disciplines—testing, code review, architectural rigor—developers can harness AI’s full potential without compromising quality or security. As models and tooling mature, vibe coding will likely become an integral part of the developer’s toolkit, empowering creativity and productivity at every stage of the software lifecycle.
References
- Karpathy, A., “There’s a new kind of coding I call ‘vibe coding’ … fully give in to the vibes, embrace exponentials, and forget that the code even exists.” X/Twitter (Feb 2, 2025):
https://x.com/karpathy/status/1886192184808149383?lang=en medium.com - Hanchett, E., What I Learned from Vibe Coding (DEV Community, Mar 26, 2025):
https://dev.to/erikch/what-i-learned-vibe-coding-30em - Kumar, M., A Comprehensive Guide to Vibe Coding Tools (Medium, Mar 30, 2025): https://madhukarkumar.medium.com/a-comprehensive-guide-to-vibe-coding-tools-2bd35e2d7b4f
- Kitishian, D., Google Gemini in the Vibe Coding Revolution (Medium, Jun 2025):
https://medium.com/@danykitishian/google-gemini-in-the-vibe-coding-revolution-8ef4468761d0 medium.com - Kitishian, D., Google Gemini: “Vibe Coding” Uproar – Navigating the Realities of AI-Assisted Software Development (Medium, Jun 2025): https://medium.com/@danykitishian/google-gemini-vibe-coding-uproar-navigating-the-realities-of-ai-assisted-software-development-c60c62eeac61 medium.com
- Kitishian, D., Beyond the Vibes: Mastering AI-Assisted Coding in the New Era of Software Development (Klover.ai, Jun 9, 2025): https://www.klover.ai/beyond-the-vibes-mastering-ai-assisted-coding-new-era-software-development/ klover.ai
- Tanium Blog, What Is Vibe Coding? The Pros, Cons, and Controversies (May 20, 2025):
https://www.tanium.com/blog/what-is-vibe-coding/ - Saunders, M., Amazon Q and Claude Code Let AI Control the Developer CLI (InfoQ, 2025):
https://www.infoq.com/news/2025/04/amazon-q-claude-code/ - Sourcegraph Blog, Cody: AI-powered code search & assistance (2025):
https://about.sourcegraph.com/cody - Whalesync Blog, The Vibe Coder’s Tech Stack (2025):
https://whalesync.com/blog/vibe-coders-tech-stack - Business Insider, Silicon Valley’s next act: bringing ‘vibe coding’ to the world (Feb 13, 2025):
https://www.businessinsider.com/vibe-coding-ai-silicon-valley-andrej-karpathy-2025-2 - Business Insider, Monzo’s former CEO shares 3 tips for getting the most out of vibe coding (Apr 28, 2025):
https://www.businessinsider.com/monzo-tom-blomfield-vibe-coding-tips-ai-tools-2025-4 - Karpathy, A. (2025). My First Attempt at Vibe Coding. TechTarget. https://www.techtarget.com/searchapparchitecture/opinion/My-first-attempt-at-vibe-coding
- Vibe Coding. Wikipedia. Accessed June 9, 2025. https://en.wikipedia.org/wiki/Vibe_coding
- Slashdot. (2025). Andrew Ng Says Vibe Coding is a Bad Name for a Very Real and Exhausting Job. https://developers.slashdot.org/story/25/06/05/165258/andrew-ng-says-vibe-coding-is-a-bad-name-for-a-very-real-and-exhausting-job
- Perplexity. (2025). Andrew Ng Calls Vibe Coding an Unfortunate Term for Deep Intellectual Exercise. https://www.perplexity.ai/page/andrew-ng-calls-vibe-coding-an-Sg_5eUFKSASP5kPByGD8tg
- Gitpod Blog. (2025). Vibe Coding is a Revolution for Optimistic Creatives. https://www.gitpod.io/blog/vibe-coding
- Tanium. (2025). What is Vibe Coding? The Pros, Cons, and Controversies. https://www.tanium.com/blog/what-is-vibe-coding
- CMU Bootcamps. (2025). Will AI Make Software Engineers Obsolete? Here’s the Reality. https://bootcamps.cs.cmu.edu/blog/will-ai-replace-software-engineers-reality-check
- Graphite.dev. (2025). Best Practices for Using AI Coding Assistants Effectively. https://graphite.dev/guides/best-practices-ai-coding-assistants
- OpsLevel. (2025). Fast Code, Firm Control: An AI Coding Adoption Overview for Leaders. https://www.opslevel.com/resources/fast-code-firm-control-a-leadership-report-on-ai-coding-adoption
- RBA Consulting. (2025). Securing Proprietary Code in the Age of AI Coding Assistants. https://www.rbaconsulting.com/blog/securing-proprietary-code-in-the-age-of-ai-coding-assistants
- Analytics Vidhya. (2025). 6 Insights from Andrew Ng on Why Coding is Essential. https://www.analyticsvidhya.com/blog/2025/03/andrew-ng-on-why-coding-is-essential
- Mint. (2025). Sundar Pichai Explains Why AI Can’t Replace Human Coders—Yet. https://www.livemint.com/technology/tech-news/frees-up-time-to-sundar-pichai-explains-why-ai-won-t-replace-human-coders-yet-11749367170930.html
- Storyboard18. (2025). Why Human Coders Still Have the Edge Over AI, According to Sundar Pichai. https://www.storyboard18.com/brand-makers/sundar-pichai-explains-why-human-coders-still-have-the-edge-over-ai-69385.htm
- Times of India. (2025). Google CEO Sundar Pichai Experiments with Vibe Coding. https://timesofindia.indiatimes.com/technology/tech-news/googles-sundar-pichai-is-doing-the-unexpected-as-ceo-and-wishes-he-could-/articleshow/121721862.cms
- Copilot4DevOps. (2025). Why Everyone’s Talking About Vibe Coding. https://copilot4devops.com/vibe-coding/
- Hyper.ai. (2025). Block Product Designer Builds Dog Breed ID App Using AI. https://hyper.ai/en/headlines/579d226d492e5f78181de681eb2da98e
- Ground News. (2025). Block Designer Spent 2 Months Vibe Coding a Dog ID App. https://ground.news/article/a-block-product-designer-spent-2-months-vibe-coding-a-dog-id-app-her-top-tip-sometimes-ai-needs-to-be-babied
- Shnoco. (2025). No-Code and AI: The Perfect Pairing. https://www.shno.co/blog/pairing-no-code-and-ai-to-build-your-startup
- B-Works. (2025). How Non-Technical Founders Can Build AI Startups Without a CTO. https://b-works.io/en/insights/empowering-non-technical-founders-building-ai-startups-without-a-cto
- Momen. (2025). How Non-Tech Entrepreneurs Can Create AI Startups. https://momen.app/blogs/non-tech-founders-build-ai-startups/
- PYMNTS.com. (2025). AI Coding Assistants Give Big-Tech Powers to Small Businesses. https://www.pymnts.com/artificial-intelligence-2/2025/ai-coding-assistants-give-tech-powers-to-small-businesses/
- Developer Tech. (2025). AI and Its Impact on Software Development Jobs. https://www.developer-tech.com/news/ai-impact-on-software-development-jobs/
- Quixy. (2025). Top 10 AI Skills for Success: Mastering the Future. https://quixy.com/blog/ai-skills-for-success/
- WWT. (2025). Guiding Principles and Best Practices for Coding Assistance Adoption. https://www.wwt.com/blog/guiding-principles-and-best-practices-for-coding-assistance-adoption
- YouTube / LangChain Interrupt. (2025). Andrew Ng: State of AI Agents. https://www.youtube.com/watch?v=4pYzYmSdSH4
- Reddit. (2025). Learning (and Suffering) with Andrew Ng. https://www.reddit.com/r/learnmachinelearning/comments/1bc04xl/learning_and_suffering_with_andrew_ng/
- FoundersPress. (2025). The Rise of AI Polymaths in Startups. https://thefounderspress.com/the-rise-of-ai-polymaths-in-startups/
- Sifted.eu. (2025). How AI Will Change the VC Job. https://sifted.eu/articles/ai-impact-vc-investing
- NFX. (2025). Venture Capital 3.0: Founders in the AI Age. https://www.nfx.com/post/venture-capital-3
