AI’s Vibe-Coding Era — How The Shift To Apps Changed The Race

AI Apps: The New Darlings of the Tech World

Key Concepts: AI apps (wrappers), foundation models, vibe coding, application layer, commoditization of models, innovator's dilemma, proprietary data, contextual reasoning engine.

The Rise of AI Apps (Wrappers)

• Definition: AI apps, also known as wrappers, are applications built on top of existing AI models (like ChatGPT, Claude, or Llama) rather than developing proprietary models from scratch.
• Funding and Valuation: AI apps like Perplexity are attracting significant funding, with Perplexity potentially doubling its valuation to $18 billion.
• Early Criticism: Initially, AI wrappers were viewed negatively as "second-rate middlemen" simply adding an interface to existing technology.
• Shift in Perception: The perception is changing as the value is seen to move "up the stack" to the application layer, where AI is used to solve specific human problems.
• Bundling and Unbundling: AI wrappers are now seen as "bundling" application intelligence and insights on top of foundation models to solve real business problems.
• Advantages: AI apps offer better user interfaces, customized features, and are increasingly building proprietary data and fine-tuning models for specific use cases.

Manus: A Case Study

• Description: Manus, a new AI agent from China, gained significant attention and was hailed as "China's next DeepSeek moment."
• Functionality: Manus can screen resumes, plan travel, and create websites based on simple prompts.
• Underlying Technology: Manus is built primarily on Anthropic's Claude model with some Alibaba's Qwen, highlighting the "wrapper" nature of the app.
• User Reception: Despite relying on existing models, Manus quickly gained a waitlist of 2 million users in just seven days.

Vibe Coding: A New Paradigm

• Definition: Vibe coding refers to using AI models to generate code, allowing individuals without extensive coding skills to develop applications.
• Efficiency: Vibe coders can accomplish the work of 10 to 100 engineers by leveraging AI-powered code generation tools like Cursor, Replit, and Devon.
• Accessibility: Vibe coding democratizes app development, making it accessible to non-technical users.
• Success Story: Anysphere, the parent company of Cursor, achieved over $100 million in annual recurring revenue in just 12 months and is reportedly in talks to raise at a valuation of nearly $10 billion.
• Future Implications: The concept of "vibe coding" may extend to other fields like law ("vibe lawyering") and accounting ("vibe accounting"), as trust in AI models increases.

Perplexity: An AI Search Engine

• Description: Perplexity is an AI search engine that allows users to choose from a variety of AI models for their queries.
• Valuation: Perplexity is valued at 170 times annualized revenue, compared to 58 times for Anthropic and 43 times for OpenAI, indicating investor confidence in its monetization potential.
• Features: Perplexity offers features like an election hub, a finance dashboard, and is developing its own browser to compete with Google Chrome.

Other Notable AI Apps

• Harvey AI: Specializes in legal analysis using OpenAI technology, quadrupling its revenue in 2024 and raising a $300 million round at a $3 billion valuation.
• Abridge: Focuses on healthcare, simplifying paperwork for doctors by converting patient conversations into notes and billing codes, raising a quarter of $1 billion in its series D.

The Commoditization of Models and the Innovator's Dilemma

• Satya Nadella's Perspective: Microsoft CEO Satya Nadella believes that AI models are becoming commoditized.
• Interchangeability: The increasing number of available models makes it difficult to differentiate based on performance alone.
• Google's Innovator's Dilemma: Google's core search business is so profitable that a pure AI chatbot risks cannibalizing its traffic and revenue.
• Startup Agility: Startups like Perplexity can quickly test and ship disruptive features without being encumbered by legacy businesses.

The Future of AI Apps

• Competition: The AI app space is becoming increasingly competitive.
• Incumbent Advantages: Megacaps have significant advantages, including large cash reserves, enterprise relationships, proprietary data, and distribution channels.
• Startup Strategies: AI apps are deepening their moats by creating specialized models and leveraging proprietary data.
• Model Ownership: Owning their own models gives AI apps more control over performance, reduces dependence on suppliers, and strengthens defensibility against competitors.

Abridge: A Deep Dive

• Mission: Abridge aims to unburden clinicians from clerical work, allowing them to focus on patient care.
• Origin: Founded in 2018, Abridge initially used models like Bert and Bio Bert before transitioning to large language models (LLMs).
• Value Proposition: Abridge solves a specific problem in healthcare by simplifying paperwork and improving workflow efficiency.
• Privacy and Trust: Abridge prioritizes privacy and trust, which are essential in the healthcare industry.
• Contextual Reasoning Engine: Abridge uses a contextual reasoning engine to orchestrate multiple models, including web-scale and fine-tuned open-source models, to create the best user experience.
• Monetization: Abridge monetizes its services by demonstrating the value it creates for healthcare providers.
• Impact: Abridge is helping to address the healthcare worker shortage by making their jobs more manageable and improving their job satisfaction.

Conclusion

AI apps are rapidly evolving from simple wrappers to sophisticated solutions that address specific needs across various industries. They are leveraging existing AI models, building proprietary data, and fine-tuning models to create unique value propositions. While competition is fierce, AI apps have the potential to disrupt established markets and transform the way we interact with technology. The rise of "vibe coding" further democratizes app development, empowering individuals without extensive technical skills to create innovative solutions. The future of AI is likely to be driven by these specialized applications that solve real-world problems and enhance human capabilities.