This episode features leading VCs Alex Adagio and Daniel Barabender dissecting the explosive Crypto AI landscape, offering critical insights into genuine opportunities versus speculative hype for investors and researchers.
Host Jeff Wilser: Setting the Stage for Crypto AI Insights
- Jeff Wilser, a journalist covering Crypto AI, introduces the podcast's focus on discerning valuable projects from the froth in the rapidly evolving decentralized AI sector.
- The discussion aims to mirror the high interest observed at the Consensus conference, where attendees eagerly seek VC perspectives on viable use cases, business models, and promising founding teams in Crypto AI.
- The episode features two distinct conversations: one with Alex Adagio of YZI Labs, known for his sober analysis, and another with Daniel Barabender of Variant Fund, who delves into AI agents and composability.
Conversation with Alex Adagio, Investment Director at YZI Labs
The Current State of Decentralized AI: An Inflection Point
- Alex Adagio describes the current Crypto AI landscape as being at an "inflection point," marked by a significant surge in new teams and projects.
- He notes a dramatic increase in deal flow, citing that at a recent Token 2049 demo day, approximately 75% of pitches were from the Crypto AI vertical.
- This signals both immense interest and potential oversaturation, requiring careful discernment from investors.
- Adagio observes that this trend indicates more builders are not only using AI to augment performance but are also actively developing new AI primitives specific to the crypto industry, such as agents for DeFi or DAOs.
- AI primitives are fundamental building blocks or core functionalities upon which more complex AI systems and applications are constructed.
Promising Use Cases at the Crypto AI Intersection
- Adagio highlights two main areas of interest: novel AI applications uniquely beneficial to crypto, and traditional AI companies leveraging crypto incentive layers.
- He points to the potential of crypto incentives to foster contributions to data and model development, something difficult to achieve in traditional Web2.
- "Having that actually is very useful and very unique to crypto which you cannot achieve with traditional means you know in web 2," Alex states, emphasizing the power of tokenized incentives.
- This model allows for global, borderless community building around projects, unifying contributors with shared economic goals.
Deconstructing the Crypto AI Tech Stack
- Data Layer:
- High-quality data remains paramount for AI. Adagio sees crypto's strength in incentivizing the creation and labeling of accurate data sets.
- He suggests that if blockchain technology had been mature earlier, large-scale, globally incentivized data labeling (e.g., for medical imaging by verified experts) would have been more feasible.
- YZI Labs is an investor in Vanna, a project focused on incentivizing user data contributions for AI model training, which Alex sees as a compelling use case.
- Model Layer:
- Crypto has played a role in pushing for more open-source AI, contrasting with some "closed" approaches from major AI labs.
- The trend is less about crypto-native teams building large models from scratch and more about "inheriting the work that has been done in Web2 and augmenting it with the tools and the technology that we have in Web3."
- This involves taking open-source models and enhancing them within crypto's incentive frameworks.
Overhyped vs. Under-Discussed: AI Agents and DeFi
- Adagio notes significant hype around AI agents and agentic frameworks (systems enabling AI agents to perform tasks), partly due to the ease of launching them and a cultural overlap with meme coins.
- He expresses skepticism about the current utility of many agents and questions the necessity of tokens for all agent-based projects, suggesting some are purely speculative.
- A less-discussed but promising area is execution-focused agents in DeFi (Decentralized Finance) – financial applications on blockchains.
- These agents could automate complex, manual tasks like yield farming, significantly improving efficiency for users. Adagio believes the "holy grail" agent that autonomously finds market alpha is unlikely due to market efficiency dynamics.
Evaluating Crypto AI Startups: Team, Product, and Sustainability
- For YZI Labs, an early-stage investor, the primary focus is on the founding team's background, motivation, and "founder-to-product fit."
- Expertise or strong interest in both crypto and AI is crucial.
- Adagio emphasizes the importance of identifying genuine builders versus "tourists" chasing hype. Openness to pivoting is also a valued trait, as initial ideas often evolve.
- "Some of the most successful investments that we've done were also results of pivots," he shares.
- Ultimately, sustainable business models that generate long-term revenue and user adoption are key, even if short-term profitability isn't immediate.
Leveraging AI in VC Investment Processes
- Alex Adagio shares how his team uses AI to enhance their investment workflow:
- Synthesizing large volumes of data from various sources (Notion pages, blog posts, founder tweets) into quick summaries using LLMs (Large Language Models) – AI trained on vast text data to understand and generate language. This boosts productivity in managing high deal flow.
- Using LLMs as research assistants to quickly understand complex technical white papers by asking specific questions about subsections.
- Actionable Insight for Researchers/Investors: These AI-driven research techniques can be adapted by anyone looking to process and understand complex information more efficiently in the Crypto AI space.
Alex Adagio's Predictions for Crypto AI
- Near-Term (6-12 months): A continued, possibly steeper, growth in the number of teams building in Crypto AI, driven by lower barriers to entry as LLMs become more powerful for coding. This will necessitate robust infrastructure for secure product development.
- Long-Term (3-5 years): The emergence of "unicorns that will be run by teams of less than 10 people," highly augmented by AI technology, achieving significant revenue or volume with minimal human staff.
Conversation with Daniel Barabender, GC and Investment Partner at Variant Fund
Crypto's Core Strengths in the AI Revolution
- Daniel Barabender frames the Crypto AI intersection by focusing on crypto's fundamental strengths:
- Aggregating talent and resources: Crypto incentives can attract compute and data contributions for AI model training, overcoming high upfront costs.
- Verification: Essential for enabling composability – the ability for different systems or agents to work together seamlessly.
- Self-custody: Crucial for AI agents that may need to manage their own funds or assets.
- Barabender's perspective is grounded in how these crypto primitives can genuinely enhance AI, rather than forcing a fit.
The Future of AI Agents: Verification and Composability
- Barabender envisions a future where AI agents perform tasks autonomously, without constant human oversight. This necessitates robust verification, especially when agents interact in complex chains.
- "If you want to chain things together right now, the way we do that is through we trust each party along a chain," he explains, highlighting the current limitations. Crypto can enable trustless interactions.
- In a scenario with multiple agents performing sub-tasks and potentially making microtransactions, crypto provides the rails for verifiability and ensuring atomic completion (all-or-nothing execution) of complex delegated tasks. This is akin to DeFi's "Lego block" composability.
Do AI Agents Truly Need Crypto?
- Addressing skepticism, Barabender concedes that if the internet's future involves agents interacting primarily with a few large, trusted entities (e.g., an OpenAI agent booking a flight with an American Airlines agent), crypto's role might be limited to efficiency gains like stablecoin payments.
- However, crypto becomes indispensable in a more permissionless internet where numerous specialized, third-party agents, potentially unknown to the user, interact and require upfront payment for services. Here, crypto's trustless nature and verifiability are key.
- Strategic Implication: The necessity of crypto for AI agents hinges on whether the future internet of agents is dominated by walled gardens or a more open, decentralized ecosystem.
Economic Ownership and Bootstrapping AI
- Barabender emphasizes "economic ownership" as a powerful mechanism for resource aggregation in AI.
- He draws an analogy to "sweat equity" in startups: tokens can represent future upside for those contributing compute or data to train AI models, solving the critical bootstrapping problem for open-source or decentralized AI initiatives that lack massive initial capital.
- This democratizes access to resources needed for AI development.
Composability: The Horizontal Path to Advanced AI
- Barabender contrasts two paths to advanced AI:
- Vertical Scaling: The Big Tech approach of throwing more compute and data at monolithic models.
- Horizontal Scaling: A crypto-aligned approach where numerous specialized AI components or agents interoperate. This composability allows for emergent intelligence from simpler, focused parts.
- He argues that horizontal scaling, fostering a permissionless ecosystem of specialized contributors, could lead to more robust and diverse AI development than centralized, vertical approaches.
The World Transformed by Agent Composability
- If true agent composability is achieved, Barabender foresees a form of "super intelligence" where software can handle almost any arbitrary task by intelligently routing it among specialized agents.
- This decentralized model offers an alternative to a future dominated by a few AI monopolies, potentially preventing the kind of value extraction and control seen in other tech sectors.
- "The alternative vision is that like the software is actually given to the people," he suggests.
Key Challenges: The Fuzzy Verifiability Problem
- A major hurdle is the "fuzzy verifiability problem": how to objectively verify that an agent has performed a real-world, often subjective, task correctly (e.g., was a data scraping job "good"?).
- Barabender is hopeful that AI itself could serve as a neutral arbiter for such tasks, rather than relying on purely cryptoeconomic solutions (like stake-based voting), which can be susceptible to manipulation or capture.
- Research Focus: Developing reliable mechanisms for verifying subjective agent outputs is critical for real-world Crypto AI applications.
Red Flags and Underappreciated Opportunities in Crypto AI
- Red Flags for Investors:
- Projects shoehorning AI into crypto concepts (or vice-versa) without genuine utility or a clear problem-solution fit.
- Crypto projects lacking product-market fit that vaguely point to "AI" as a future panacea.
- Underappreciated Thesis: Data Layers Reimagined
- Barabender is excited by new approaches to data layers (shared data infrastructure for applications).
- Past attempts often failed by burdening users. The new model, exemplified by projects like Plastic (a Variant investment), involves first providing value to applications (e.g., an SDK – Software Development Kit – to help apps understand their users) and then using this to bootstrap a shared data layer. This inverts the traditional approach and could solve the cold-start problem for data ecosystems.
A Wild Future: The Triumph of Modularity
- Barabender's bold prediction for the next 10 years: the modular, composable approach to AI will ultimately win out over monolithic systems.
- He believes the power of permissionless innovation and network effects from a diverse ecosystem of contributors is currently underestimated and will scale intelligence in unforeseen ways.
- "I think that we are completely underestimating the power of permissionlessness and anybody being able to contribute their ideas and how much faster that can scale versus just the vertical approach."
Host Jeff Wilser: Concluding Thoughts and Developments
- Wilser thanks the guests for their deep dives into the Crypto AI space.
- He notes a relevant development: Vanna, a project focused on user-owned data for user-owned AI, recently announced a collaboration with Flower Labs to build a user-owned foundational AI model, signaling ongoing innovation in the decentralized AI ecosystem.
Conclusion: Navigating the Crypto AI Frontier
This episode underscores that while Crypto AI is rife with hype, genuine innovation lies in leveraging crypto's core strengths—incentives, verification, and composability—to solve real AI challenges. Investors and researchers must critically assess projects for true utility, sustainable models, and solutions to complex issues like fuzzy verifiability, while keeping an eye on the transformative potential of modular, user-centric AI systems.
