The 7 Most Powerful Moats For AI Startups

Key Concepts

• Moats: Defensive strategies to protect a business from competition.
• Seven Powers (Moats): Process Power, Cornered Resource, Switching Costs, Counter-Positioning, Brand, Network Economies, Scale Economies.
• AI Agents: Software programs that use AI to perform tasks autonomously.
• Speed: Rapid execution and iteration as a competitive advantage.
• Forward Deployed Engineering Teams (FTE): Startups acting as engineering teams for larger labs or enterprises.
• Context Engineering: Designing prompts and interactions to optimize AI model performance.
• Evals: Evaluation metrics and processes for improving AI model accuracy and reliability.
• Vertical AI SAS: AI-powered software solutions tailored to specific industries.

Moats: Defending Against Infinite Competition

The discussion centers on the concept of "moats" as defensive strategies for startups, particularly in the age of AI. The core idea is that without a moat, a business is subject to infinite competition, driving margins to zero and threatening its survival. The conversation highlights how the focus on moats has intensified among aspiring startup founders due to the rise of AI and the perception that AI-driven businesses are easily cloned.

The Seven Powers (Moats) Framework

The framework is based on Hamilton Helmer's book "7 Powers: The Foundations of Business Strategy". The discussion reinterprets "powers" as "moats," emphasizing their defensive nature. While the book's examples are dated (e.g., Oracle, Facebook, Netflix), the underlying principles remain relevant.

1. Process Power

• Definition: Building a complex, difficult-to-replicate business through intricate processes.
• Classic Example: Toyota assembly line.
• AI Agent Version: A finely-tuned AI agent honed over years to perform reliably in real-world conditions.
• Examples:
  • Case Text: Legal research platform.
  • Greenlight: KYC (Know Your Customer) AI agent for banks.
  • Casa: Loan origination AI agent for banks.
  • Plaid: Aggregates financial data from thousands of institutions.
• Key Point: The "hackathon version" of an AI agent is insufficient; defensibility lies in the painstaking engineering required for mission-critical reliability.
• Modern Application: Using codegen tools to rapidly integrate new financial institutions (as Plaid might do).
• Connection to Existing SAS: Similar to how Stripe, Rippling, and Gusto are defensible due to their complex backend logic.
• Shle Blindness: The tedious nature of perfecting AI tools for specialized verticals like KYC can deter competitors.

2. Cornered Resource

• Definition: Controlling a scarce, valuable asset or having preferential access to it.
• Characteristics: Coveted, non-arbitrageable, independently valuable, and offering preferential access.
• Classic Example: Pharma companies with patents and regulatory approvals.
• Modern Examples:
  • Scale AI & Palantir: Working closely with the DoD, building specialized data centers (skiffs), and becoming embedded in government processes.
  • Character AI: Fine-tuning LLMs to reduce serving costs by 10x.
• Startup Application:
  • Forward Deployed Engineering (FTE): Gaining access to real data and workflows by working closely with customers.
  • Developing a proprietary model: Creating a model that performs specific tasks better than general-purpose models.
• Threat: Labs restricting access to their models, treating them as a cornered resource.
• Context Engineering as a Substitute: Even if perfect AI requires extensive pre-training and fine-tuning, context engineering can achieve 80-90% of the desired results, sufficient for early-stage startups.

3. Switching Costs

• Definition: Making it expensive or difficult for customers to switch to a competitor.
• Classic Examples:
  • Oracle: Migrating data from a database is a major undertaking.
  • Salesforce: Retraining sales teams and migrating customer records is costly.
• AI-Driven Switching Costs:
  • Customized Workflows: Lengthy onboarding processes and deep customizations of AI agent logic for specific enterprise operations.
  • Memory: AI agents that remember user preferences and history create a switching cost for consumers.
• Examples:
  • Happy Robot: Integrating deeply into DHL's logistics operations.
  • Salient: Building AI voice agents with custom workflows for different banks.
• AI Reducing Switching Costs: LLMs can automate data migration and extraction from legacy systems, potentially lowering switching costs in some cases.

4. Counter-Positioning

• Definition: Doing something that incumbents cannot easily copy because it would cannibalize their existing business.
• Example: New AI native companies vs. existing SAS incumbents building AI agents.
• Pricing Model Achilles Heel: Incumbents charging per seat (per employee) may see revenue decline as AI agents automate tasks and reduce the need for employees.
• Alternative Pricing: Startups pricing based on work delivered or tasks completed.
• Engineering Culture Challenge: Incumbents struggle to adopt AI-native engineering practices, hindering their ability to deliver effective AI products.
• Vertical AI SAS Advantage: AI startups can tap into new areas of customer spend beyond traditional software budgets (e.g., customer support for HVAC companies).
• Workforce Displacement Nuance: AI can automate undesirable jobs with high attrition rates, leading to more engaging roles for remaining employees (e.g., managing AI agents instead of handling routine customer support).
• Second Mover Advantage: Being a second mover in a space allows companies to learn from the mistakes of early winners and build a better product.
• Examples:
  • Legora vs. Harvey: Legora focused on the application layer after Harvey prioritized fine-tuning.
  • Giga ML: Offering a customer support solution that works better out of the box, leading to faster onboarding.
  • Dualingo vs. Speak: Speak focuses on actual language learning through voice interaction, while Dualingo is perceived as a gamified app.
• Brand as Counter-Positioning: Building a strong brand can create a moat, even with equivalent products (e.g., OpenAI's ChatGPT vs. Google's Gemini).

5. Network Economies

• Definition: The value of a product increases as more users join the network.
• Classic Examples:
  • Facebook: More valuable as more friends join.
  • Visa: More valuable as more merchants accept it.
• AI-Era Network Effects: Data becomes the key network effect.
• Examples:
  • Foundation Model Companies: Training models on vast amounts of user data (e.g., ChatGPT feeding chat history into future models).
  • Cursor: Using every mouse click and keystroke to train its autocomplete model.
  • Salient & Happy Robot: Improving workflows with private data from enterprise customers.
• Evals as a Flywheel: Using evaluation metrics to iterate and improve context engineering based on user data.

6. Scale Economies

• Definition: Achieving lower costs per unit by investing in large-scale infrastructure.
• Classic Examples:
  • UPS, FedEx, Amazon: Massive delivery networks.
• AI Application: Primarily at the model layer, where training state-of-the-art LLMs requires significant capital investment.
• Deepseek Announcement: Potentially diminishing the power of scale economies by making it cheaper to train frontier LLMs.
• Startup Example:
  • Exa: Crawling a large portion of the web to provide search for AI agents.
• Channel 3 and Orange Slice: Crawling a big chunk of the web, have a big like static crawl on their own servers, and then have agents that run on top of those of that crawl.

Speed: The Overlooked Moat

Varun from Windsorf emphasizes "speed" as the primary moat for early-stage startups. This is not one of the seven powers in the book, but it is crucial. Cursor's one-day sprint cycles exemplify this.

When to Think About Moats

• Early Stage: Focus on finding a real problem and solving it.
• Later Stage: Once you have something valuable to defend, start thinking about moats.
• Don't Overthink It: Avoid using moat analysis to prematurely dismiss startup ideas.

Conclusion

The discussion provides a comprehensive overview of moats as defensive strategies for startups, particularly in the context of AI. While the seven powers framework offers valuable insights, the importance of speed and execution cannot be overstated. Early-stage startups should prioritize solving real problems and achieving product-market fit before focusing on long-term defensibility. The rise of AI presents both challenges and opportunities for building moats, with data, customized workflows, and counter-positioning emerging as key strategies.