Andrew Chen's Litmus Test for AI Startups
Andreessen Horowitz general partner Andrew Chen shared a key question for evaluating AI startups in 2026: "Can your product survive 10x base model improvements in 18 months?" The query emphasizes the need for startups to build defensible moats beyond simply leveraging a powerful third-party AI model, which will rapidly commoditize.
- Before joining Andreessen Horowitz, Andrew Chen led rider growth teams at Uber and has authored a book on network effects called *The Cold Start Problem*. His work focuses on how startups can build value and defensibility through user networks, a concept highly relevant to creating moats in the AI era. - The "10x improvement" is a significant benchmark, as many experts in 2026 note that relying solely on a superior model is not a defensible long-term strategy due to the rapid commoditization of AI. Competitors, including open-source alternatives, often close the quality gap on core model performance quickly. - Defensible moats in the current AI landscape often revolve around proprietary data and workflow integration. A key strategy is to build products that generate unique data through user interaction, creating a feedback loop that continuously improves the product in a way competitors cannot replicate. - Embedding AI tools deeply into a customer's daily operations creates high switching costs. This "workflow embedding" means that even if a competitor has a slightly better model, the cost and effort of retraining teams and reconfiguring processes make switching impractical. - Other non-functional advantages are becoming critical for defensibility in 2026, including brand recognition, team execution speed, and building a strong ecosystem around a product. A team's ability to iterate and ship daily can create an operational tempo that larger incumbents cannot match. - Chen is heavily involved in a16z's SPEEDRUN accelerator, a program that invests up to $1 million in early-stage startups and provides access to the firm's network. The program, which runs in San Francisco, has funded over 150 companies since its launch in 2023, with a significant focus on AI. - Examples of a16z SPEEDRUN participants building defensible AI businesses include General Magic, which creates AI agents for the insurance industry, and Loops AI, which is developing autonomous agents for e-commerce. These companies focus on specific vertical applications and deep workflow integration rather than just foundational model development. - The current venture capital landscape in 2026 shows a shift in focus from "AI for everything" to more specialized, niche AI tools with a clear path to profitability. Investors are now looking for sustainable business models and tangible metrics, a sentiment that aligns with Chen's call for durable moats.
