20VC: Horizontal vs Vertical Agents: Who Wins | Model Routing: Commodity or Increasing in Value | The Future of the Inference Market | Why Founders Should Take Discounts for Tier 1 Firms with Harrison Chase, LangChain

Building reliable AI agents remains a significant technical challenge due to the non-deterministic nature of large language models and the complexity of multi-step workflows. Harrison Chase, founder of LangChain, argues that the most effective architecture for agents involves a "harness" layer that manages state, memory, and tool interaction, rather than relying solely on the model itself. This harness layer enables portability, allowing developers to switch between models while maintaining control over proprietary context and domain-specific logic. While model routing is often debated, it is best implemented within the harness to leverage contextual awareness. As AI adoption scales, enterprises must prioritize accuracy and cost-efficiency through visibility and rigorous evaluation, moving beyond simple model swapping to sophisticated, domain-specific agent architectures that can learn and improve over time.