Navigating the Gen AI investment landscape: wisdom from VC experts

On Day 3 of Launchable, Tim Porter of Madrona was invited to moderate an investor panel focused on the current fundraising climate. Joining the discussion were Yohei Nakajima from Untapped Capital and founder of Baby AGI, Rama Sekhar of Norwest Venture Partners, and Kanu Gulati representing Khosla Ventures. Read some of their valuable insights below:

‍Tim: What's your current Gen AI investment thesis and evaluation framework? 

‍Rama: In Gen AI, I see a three-layer hamburger [stack] that’s emerged. At the bottom are foundation models, which we haven’t been investing in. At the top are applications, ranging from thin wrappers on APIs to complex co-pilots built around private LLMs. Then there's the middle layer, “enabling technologies" such as capabilities around model security and trust and safety that enable enterprises to deploy these models. My primary focus is on this middle layer with some interest in the application layer.

‍Yohei: At the pre-seed stage, we're focused on teams. We look for teams that can build and improve their product rapidly and stay up to date with emerging trends in Gen AI.

‍Kanu: At Khosla Ventures, we invest in early, bold, impactful Gen AI solutions. We consider both applications and the foundational technology stack. We assess whether they are augmenting human tasks, automating them, or serving as co-pilots. We also consider the team's mix of domain knowledge and technical expertise and how they are building integrations that impact the product's success.

‍Tim: Let’s chat about creating moats for a Gen AI startup. How do you build something defensible in this space when things are moving this quickly?

‍Rama: Creating a moat in Gen AI is a complex challenge. Data is often considered a moat, but it may not apply to early-stage companies. Technology, workflow, and product experience can be sources of differentiation. The type of moat varies depending on the application they are building.

‍Yohei: Building and iterating quickly is a competitive advantage. Creating a feedback loop and continually improving the product based on user engagement can also be a moat.

‍Kanu: Focus on product experience, proprietary data, an efficient feedback loop, and consider the depth of integration, where shallow integration demands exceptional product quality and momentum, while deep integration involves tightly coupling with workflows and data collection, contributing to a strong moat for your Gen AI startup.

‍Tim: Kanu, you mentioned shallow versus deep integration, which is intriguing. Is the data flywheel the primary focus, or just a part of the bigger picture?

‍Kanu: It's an important piece of the puzzle. Another key aspect is defining the desired product experience and identifying the core components to address from the beginning. We should also explore access to proprietary data and efficient data collection methods before or during the product's production. The approach to working with LLMs varies, including prompt engineering, fine-tuning custom models, or building new ones. We need to determine which approach adds value to a specific application and assess potential advantages as we progress to more advanced LLM iterations like GPT-4 and beyond. The framework for risk mitigation is a top priority in this context.

‍Tim: Could you share some specific examples of Gen AI companies you've invested in and what excited you about them?

‍Yohei: Absolutely, one of our exciting investments is in Wokelo AI, a company that specializes in AI-generated business research. What really piqued our interest was the fact that I had actually been personally working on a similar tool in-house. We were somewhat known for our AI analyst memos, and Wokelo AI came to our attention because they were developing something akin to what we were building. It's fascinating because it's a product I was personally involved in, and we've had a real hands-on experience with it.

‍Kanu: Let me highlight Replit, a company specializing in copilot and autopilot solutions for developers. What's particularly riveting about Replit is their exceptional traction. They aim to empower developers of all skill levels, making everyone better at programming. When we invested, they had already amassed over 20 million users, and they continue to grow.

‍Rama: We invested in Lumix, an Israeli company emerging from stealth mode. They've taken Gen AI beyond the realm of traditional software and into the physical security sector. Imagine enhancing your existing security cameras with innovative capabilities, from measuring retail spaces to detecting workplace hazards. It's a groundbreaking concept, pushing Gen AI beyond its conventional software boundaries and into the physical world.

‍Tim: How do you assess whether Gen AI companies are developing the right product interface, and how do you distinguish them from potential platform features?

‍Kanu: Distinguishing between if a product can become a company or is  a feature is challenging. We have open discussions with founders to understand their long-term vision and adaptability to tech changes.

‍Tim: Any advice for founders pitching Gen AI businesses?

‍Rama: Be authentic about where AI is genuinely used, and communicate the impact your solution can have in the market.

‍Yohei: Prioritize getting feedback and choose a demo that resonates with your target audience.

‍Kanu: Focus on your market, demonstrate what you've built, and clarify your vision, risk management plan, and future goals.

Special thanks to Tim and panelists.