How Anthropic’s $100M Anthology Fund Works | Menlo Ventures

Founder Investment Philosophy

• Thesis vs. Founder Selection: Almost no iconic venture-backed companies (e.g., OpenAI, Facebook) originated from a pre-existing VC thesis; investments were made based on founder competence and the potential value of the work, not a specific predicted trend like "LLM chatbots."
• Motivation Filter: The primary screening metric for founders is determining if their drive is authentic rather than driven by the perceived status and "coolness" of being a founder.
• Long-term Conviction: Investors look for founders who, five years post-funding, would still choose to work on the mission regardless of the company's immediate success or failure.
• Cult of Personality: A critical bottleneck is a founder's ability to act as a "cult leader" to hire and retain talent; many smart founders lack the network or charisma to secure a team.
• Relationship Building: Conviction is difficult to establish without pre-existing relationships, as founders often sell aggressively during pitches; Menlo emphasizes building trust with founders before funding rounds begin.

Specific Investment Focus

• Goodfire: A portfolio company specializing in mechanistic interpretability ("brain surgery for AI models") to explain black-box AI decisions, distinct from empirical evaluation methods.
• Labor Market Gaps: Significant investment opportunities exist in industries facing hiring shortages (e.g., insurance, logistics, trucking, tax) where AI must solve immediate labor gaps rather than just theoretical problems.
• Demographic Shifts: Long-term bets are considered regarding the economic repercussions of declining global birth rates and population shrinkage on consumption patterns.
• Tasteful Product Building: Preference for "beautiful" products that prioritize user experience over forced AI integration, with a strong belief that Product-Led Growth (PLG) models will disrupt enterprise sales cycles due to viral adoption.
• Technical Risk Underwriting: Support for teams conducting fundamental research where academic institutions no longer have the capital to fund frontier science.

AI Infrastructure & Evolution

• Current Cycle Stage: The industry has moved from scaling on internet data (GPT-3 era) and Reinforcement Learning from Human Feedback (RLHF) to "RL World," where proprietary, high-quality data is purchased to train models on specific domain expertise.
• Bottlenecks: Reinforcement Learning is limited by the manual nature of data acquisition and the inefficiency of learning from single-endpoint rewards; scaling laws for RL are less intuitive than for pre-training.
• Future Directions: Key research areas include improving data efficiency (learning from fewer samples), expanding test-time compute (agentic workflows), and enhancing model reliability/determinism (memory and structural understanding).
• Economic Tiring Test: A key metric for AI progress is whether an AI can perform a task for a set monetary amount that a human would charge, with the goal of pushing that cost down.

Funding & Market Dynamics

• Seed Check Size: Seed investments over $100M are viewed as difficult to justify from a standard fund unless structured as SPVs where the VC assumes minimal risk.
• Over-financing Risks: Excessive capital can lead to an "illusion of success," reduced hunger, the "50 lunch problem" (fancy spending), and recruitment difficulties regarding equity valuation.
• Valuation Premiums: AI-enabled startups command a ~30% premium, but this is only justifiable for companies with deep technical expertise and a clear, economically viable application, not just for buzzword compliance.
• Hiring Incentives: Raising at inflated valuations can help compete with tech giants (like Anthropic) by offering higher paper equity value to employees, though this is often a temporary strategy.

Labor Market & AI Impact

• Efficiency vs. Displacement: Current tech layoffs are driven by a desire for efficiency and the elimination of low-value roles rather than purely AI-driven replacement, though AI will eventually augment or replace more white-collar tasks.
• Historical Context: While new technologies displace old jobs, they historically create new categories of employment; the current transition may cause short-term pain but long-term reallocation.
• Job Creation: There is a belief that AI will create jobs that allow humans to work more effectively (higher "firepower") rather than solely replacing them.

Personal Insights & Methodology

• Public Writing Strategy: The host's success on X (Twitter) is attributed to a strict rule of only posting content that is helpful to others, avoiding political commentary, and focusing on data-driven facts.
• Data Visualization: Preferred tools for creating charts include Python (via custom scripts for Claude), Google Slides, and Apple's Freeform app for assembling visual narratives.
• Admiration for Arvind Gupta: Cites the Glean CEO as a primary role model for his Japanese philosophy of respecting the craft, rejecting ego/status symbols, and viewing hard work as a gift.
• Focus Discipline: Key lesson from Arvind is to avoid overthinking defensibility or scaling prematurely; the only question that matters at the early stage is "Do customers love this product?"
• Acquisition Mindset: A founder's willingness to reject a massive acquisition offer to continue running a "cool company" indicates a deeper commitment to the mission than simply seeking liquidity.