Mar 13, 2025, 4:28 PM ET

Vibe Coding is Here - How AI is Changing How We Build Online

Vibe coding is defined as providing an LLM with natural language instructions (e.g., "this vibes well with me") and allowing the agent to execute, iterate, and deploy code with minimal human intervention.
The trend, popularized by Andrej Karpathy, involves users ranging from non-technical individuals to experienced engineers who prioritize rapid iteration over writing code from scratch.
The term implies a shift where the primary interaction is natural language feedback rather than explicit syntax writing.

IDE-Integrated Tools: Platforms like Cursor target developers by integrating agents directly into code editors, facilitating code generation and editing within a familiar environment.
Browser-Based "Text-to-App" Tools: Newer entrants (e.g., Replit Agent, Lovable, Bolt, v0 from Vercel) operate in the browser, generating visually distinct web applications via text prompts.
User Segmentation:
- Engineers prefer IDEs (Cursor) for full control and complex development workflows.
- Non-technical users and creatives favor browser tools for their ability to generate aesthetically pleasing, dynamic assets quickly.
- Agencies and consultants are adopting these tools to take on lower-margin projects that previously lacked economic viability due to development time.

Rapid Revenue Growth: Emerging companies like Bolt and Lovable have reported reaching $20 million ARR in two months (or $10 million ARR in two months), indicating massive latent demand.
Democratization of Software: The trend has enabled the creation of "Software for One," allowing individuals to build highly personalized apps (e.g., tracking plant watering, personal book sales pages) that were previously inaccessible without professional developers.
Shift from Static to Dynamic: Unlike legacy no-code tools (Wix, Squarespace) that often produce static sites, vibe coding tools generate functional code capable of handling databases, APIs (e.g., OpenAI), and authentication (e.g., Clerk, Google Auth).

Model Capabilities: Current LLMs excel at web development due to the prevalence of JavaScript/TypeScript data on the internet and the maturity of frameworks like React and Next.js.
Context Management Issues: Tools often struggle with context loss as codebases grow; a single requested change can break previously working components (e.g., an image insertion breaking authentication).
Optimism Bias: LLMs exhibit high confidence even when stuck in loops, leading to frustrating user experiences where agents fail to pivot despite repeated prompting.
Verification Gaps: The industry lacks robust, automated testing and verification layers; future iterations are expected to focus heavily on unit testing and state verification rather than just code generation.

Current Pricing: Most companies utilize token-based consumption models (free tiers followed by monthly subscriptions based on usage), which can create cost unpredictability for users.
Future Pricing Direction: There is a market shift toward value-based pricing (aligning fees with user revenue) or per-seat models for enterprise collaboration features.
Market Expansion: The category is not displacing existing markets but expanding the total addressable market (TAM) to include the ~99% of the population that previously could not build with code.

Specialization: The market will likely segment further based on user technical sophistication, with interfaces tailored specifically for consumers (simplified) versus engineers (granular control).
Design Integration: Next-gen tools are expected to move beyond text prompting to include Figma-like interfaces for direct visual manipulation and "vibe board" inputs (uploading screenshots for design generation).
Extensibility (MCP): The emergence of the Model Context Protocol (MCP) allows agents to be extended with custom tools (e.g., sending emails, accessing local files), functioning as a next-generation API layer.
Incumbent Defense: Startups maintain defensibility against giants like Google and Figma due to the incumbents' legacy architecture; AI-native startups can innovate faster by building entirely new workflows rather than retrofitting old ones.
Bi-Directional Interaction: Future agents will likely become more vocal, suggesting design options and asking clarifying questions rather than simply executing static prompts.