Generative AI Moats in B2B with Emergence Capital’s Jake Saper

The episode begins with a comprehensive overview of what large language models are, emphasizing their ability to understand and generate human language using deep learning. Jake explains that models like GPT (Generative Pre-Trained Transformer) are based on transformer architecture, introduced around 2017, enabling them to process data in parallel and produce highly realistic and human-like text. A significant point discussed is the pre-trained nature of GPT models, trained on data up to 2021, highlighting their limitations in real-time knowledge and the ongoing efforts to update or augment these models with current data. The conversation also touches on concerns about recursive training data and the impact on the quality of outputs, emphasizing the importance of human contribution and insight to maintain originality and value in AI-generated content.

This section delves into how companies can leverage generative AI for building defensible businesses. Jake introduces a framework based on the importance of accuracy and real-world outcomes, illustrating that different use cases require different levels of precision. For instance, consumer-focused chatbots where accuracy is less critical versus B2B applications like legal contracts or healthcare, where factual correctness is vital. He discusses the concept of human-in-the-loop systems, where humans verify and add insights, creating feedback loops that improve AI over time. Defensibility often hinges on proprietary data, insights, and outcomes that cannot easily be replicated. The discussion emphasizes that successful AI applications will need to incorporate careful scaffolding, workflows, and ongoing data collection to sustain competitive advantages.

The episode explores how user interfaces and experience design are evolving with AI. Jake notes that chat interfaces are becoming dominant, but hybrid solutions combining conversational and traditional UIs will likely emerge. The importance of keeping users actively engaged and integrating AI suggestions into workflows, rather than replacing critical human judgment entirely, is stressed. Examples like Notion's AI features demonstrate how embedding AI within existing platforms can improve productivity. The conversation underscores that AI should serve as a co-pilot, amplifying human effort rather than automating away all roles. Challenges include avoiding complacency, maintaining human oversight, and designing interfaces that sustain user attention and input.

Jake discusses how AI is transforming industry landscapes and how incumbents will face differing levels of disruption. Sectors reliant on old UI paradigms, like traditional point-and-click enterprise software, may struggle due to the innovator’s dilemma. Startups exploiting generative AI for niche jobs—such as legal document review or pricing strategy—can create new categories and significant value. The episode highlights the importance of proprietary data and outcomes, which underpin defensibility, especially as off-the-shelf models become more accessible. Jake warns of the risk of ‘AI hammers looking for nails’—many startups superficial in their use of AI—making differentiation crucial. He also notes the critical role of compliance, privacy, and regulatory considerations, as well as the possibility of future regulations shaping how AI can be used.

Attention is given to how AI technology impacts traditional SaaS pricing models. Jake advocates shifting away from per-seat pricing, which may be inefficient and inhibit widespread adoption, towards value-based models aligned with outcomes and customer ROI. He explains that integrating proprietary outcome data can create long-term moats, allowing companies to monetize the real value they generate. Novel pricing strategies, such as charging based on expected use or business outcomes, could evolve, incentivizing users to share data and improve AI performance. The discussion also covers the impact of variable infrastructure costs and how to pass those costs to customers without discouraging adoption, emphasizing that pricing must adapt as AI becomes embedded in business workflows.

The final sections reflect on the challenging macroeconomic environment for venture investing, characterized by higher interest rates, market volatility, and recent bank crises. Despite these headwinds, Jake emphasizes the importance of disciplined, long-term investment strategies and maintaining focus on value creation. He notes that the AI boom’s momentum continues to excite LPs and investors who recognize the potential for durable, AI-enabled startups. The conversation covers the importance of brand and reputation, patience in deploying capital, and staying committed to pioneering AI innovations despite short-term market dislocation. Jake also warns about potential risks of overhyped startups operating without real defensibility, but remains optimistic about the opportunities for those who build with genuine value and sustainability in mind.