Startups You Can Now Build With AI

Key Concepts

• AI-powered marketplaces
• Three-sided marketplaces
• Personalized learning
• AI agents
• Tech-enabled services
• Gross margins
• MLOps
• Platform neutrality
• Innovator's dilemma
• System prompt vs. User prompt

Recruiting Startups and AI-Powered Marketplaces

The discussion begins with the observation that recent advancements in AI, particularly Large Language Models (LLMs), have unlocked new possibilities for startup ideas, especially those that previously struggled. One specific area highlighted is recruiting startups.

• Triple Bite Example: The experience of Triple Bite, a recruiting startup founded around 2015, is used as a case study. Triple Bite aimed to create a curated marketplace for engineers, but the evaluation process required extensive manual technical interviews and years of building a labeled dataset for machine learning.
• AI's Impact: AI, specifically code generation models, now enables instant code evaluation, eliminating the need for years of data collection.
• Meror as an Example: Meror, a current AI startup, is presented as an example of a company leveraging LLMs to evaluate software engineers from day one, allowing them to expand into other knowledge work more easily.
• Three-Sided Marketplaces: The conversation highlights the potential of transforming three- or four-sided marketplaces into simpler two-sided models using AI. For example, Duolingo potentially using AI for language partners.
• Psychological Element: The psychological challenge of entering a space where previous startups have failed is discussed. Founders need to overcome skepticism and investor burnout, even with the "LLMs change everything" narrative.
• Instacart Analogy: The Instacart story is used as an analogy. Just as mobile phones enabled Instacart to succeed where Webvan failed, LLMs are now enabling new possibilities in various industries.

AI in Education and Personalized Learning

The discussion shifts to the education sector, focusing on personalized learning and AI-powered tools for teachers and students.

• Hyperpersonalization: Hyperpersonalization is identified as a "holy grail" for edtech companies, made possible by AI.
• Personalized Tutor: The idea of a personalized AI tutor in everyone's pocket is presented as a long-standing dream now within reach.
• Revision Dojo: Revision Dojo, an exam prep tool funded by YC, is mentioned as an example of a successful application of AI in education, offering tailored flashcards and engaging content.
• Adexia: Adexia, another company, provides AI tools for teachers to grade assignments, addressing a major pain point that contributes to teacher burnout.
• Private vs. Public Schools: The discussion touches on the disparity between private and public schools in adopting AI-powered educational tools, raising questions about policy changes needed to support public schools.
• Distribution Challenges: The challenge of distribution for AI-powered learning apps is discussed. Better products don't automatically guarantee wider adoption.
• Cost of Intelligence: The decreasing cost of AI is highlighted, suggesting a potential shift towards freemium models where basic intelligence is free, and users pay for advanced features.
• Speak Example: Speak, a language learning app, is presented as a success story, demonstrating the potential of personalized language learning powered by AI.

Business Models and Moats in the AI Era

The conversation explores business models and competitive advantages (moats) for AI startups.

• Enterprise vs. Consumer Budgets: The difference in budget allocation between enterprises and consumers is discussed. Companies are willing to pay significantly more for AI solutions that replace entire teams (e.g., customer support, analytics).
• Business Model Transformation: AI can transform the business model of a product. For example, a personalized learning app that rivals a human tutor can command a much higher price point.
• Moats: The importance of building moats, such as brand recognition, switching costs, and integration with existing technologies, is emphasized.
• OpenAI's Role: The role of OpenAI is discussed, with the speakers suggesting that OpenAI is not necessarily trying to eliminate startups but rather wants them to succeed using its API.
• Platform Neutrality: The need for platform neutrality is raised, drawing parallels to net neutrality and the Windows browser choice case. Users should have the freedom to choose their preferred AI assistants (e.g., Siri, Google Assistant) without being locked into a specific platform.

Google vs. OpenAI and the Innovator's Dilemma

The discussion delves into the competition between Google and OpenAI, highlighting Google's potential advantages and its challenges in fully leveraging its AI capabilities.

• Gemini's Performance: Despite having access to vast user base, Gemini's consumer usage is significantly lower than ChatGPT's.
• Intangible Moat: The intangible moat of being the first in a space and establishing a reputation as the best product is discussed.
• Google's Internal Structure: Google's internal structure and competing orgs are cited as potential reasons for its struggles in effectively integrating AI into its products.
• TPUs as Dragons: Google's TPUs (Tensor Processing Units) are metaphorically referred to as "dragons," giving the company a potential cost advantage in AI development.
• Innovator's Dilemma: The innovator's dilemma is presented as a challenge for Google. Replacing Google Search with an AI-powered chatbot could potentially cannibalize its existing revenue streams.
• Meta's AI Integration: Meta's approach to AI integration in WhatsApp is criticized for being invasive and lacking design taste.
• System Prompt Control: The importance of allowing users to customize the system prompt in AI applications is highlighted, referencing Pete Cuman's blog post on Gemini's Gmail integration.

Tech-Enabled Services 2.0

The conversation revisits the concept of tech-enabled services, arguing that AI now makes it possible to build full-stack companies with software-like margins.

• Tech-Enabled Services Wave: The tech-enabled services wave of the 2010s is discussed, with Triple Bite and Atrium as examples.
• Gross Margin Issues: The main reason for the failure of many tech-enabled services companies was their low gross margins due to high operational costs.
• Full Stack Companies with AI: AI agents can now automate many of the tasks previously performed by human operators, enabling full-stack companies to achieve software-like margins.
• Legora Example: Legora, a YC-funded company building AI tools for lawyers, is presented as an example of a company that could potentially become the "biggest law firm on the planet" by leveraging AI agents.
• Virtual Assistant Marketplaces: The potential for AI-powered virtual assistants to handle knowledge work and even hire real people is discussed.

MLOps and the Importance of Curiosity

The discussion touches on the evolution of MLOps and the importance of following one's curiosity in the AI era.

• MLOps in 2020: The speakers recall a time when YC was not interested in funding MLOps companies due to a lack of demand for ML tooling.
• Replicate and Olama: Replicate and Olama are mentioned as examples of companies that persevered in the MLOps space and eventually found success when the technology caught up.
• Deepgram: Deepgram, a speech-to-text company founded by physics PhDs, is presented as another example of a company that initially struggled but eventually thrived due to the rise of voice agents.
• Follow Your Curiosity: The importance of following one's curiosity and exploring interesting technologies is emphasized as a key to finding successful startup ideas in the AI era.
• Magical Output: The potential to achieve "magical output" by combining the right prompts, datasets, ingenuity, and taste is highlighted.
• Lack of Internal Transformation: The speakers note that many established companies are not yet fully embracing AI and transforming their internal processes.

Conclusion

The main takeaway is that there has never been a better time to build, as AI has unlocked countless new possibilities. The best way to find these opportunities is to follow one's curiosity and keep building.