E452: Why 95% of AI Projects Fail, Anthropic’s Valuation, and Why AI Is Still in the First Inning

The real moat in AI sits behind firewalls

Enterprise AI is still barely tapped, and most deployments fail. Joseph Coyne's answer is to back vertical companies with an unfair industry edge and proprietary data behind firewalls that LLMs cannot reach, since that asset compounds and cannot be bought. The venture math has shifted, with agentic coding letting technical founders do more with less while Series A and B rounds price like the D and E rounds of the past.

Takeaways

• Most enterprise AI deployments fail. Coyne cites Deloitte's finding that 95% of enterprise AI deployments fail. Survivors are deeply technical organizations with strong data infrastructure and security. Legacy firms must fix unruly, siloed data and inconsistent tagging first, and adoption has to be top-down because most leaders are not AI-native.
• AI adoption is still in the first inning. Adoption moves through S-curves that start with the world's 50m developers before reaching enterprises and consumers, with agents as a possible fourth wave. Coyne says enterprise AI is only scratching the surface, and Studio VC now sees every company arrive with an AI strategy.
• Data behind firewalls is the durable moat. About 80% of the world's data sits behind firewalls LLMs cannot reach. Companies that earn customer trust and bring subject matter expertise build an asset rivals cannot buy, and the data compounds, as Tesla's self-driving data shows. Weisburd calls the data moat AI's most sustainable advantage.
• Anthropic's enterprise focus earns a valuation boost. Coyne gives Anthropic a boost over OpenAI because Claude Code makes it more enterprise-focused while OpenAI skews consumer. He calls the strip-three-zeros valuation argument fair but says the outcome is long-tail, with low switching costs and $0 to $10 monthly consumer willingness to pay.
• Agentic coding makes technical founders cash efficient. Advisors at Meta and Amazon report doing four times as much work in half the time. Series A and B rounds are priced where D and E rounds used to be, and Coyne resists $100m pre-seed prices because outcomes are low probability.
• Vertical AI is won on unfair industry edges. Coyne looks for an unfair fight, asking who knows the industry best and who has the license to win. Studio VC leans into companies that protect against AI's dark side, and Bark flags self-harm and predators from about 7m kids' data points.

Selling AI into the enterprise

1. Do the unscalable work first. Consultative pilots teach the startup the enterprise's real problems, and the best AI companies pair that with an enterprise software workflow that unlocks data and runs AI securely.
2. Replicate with agentic workflows. Coyne says the next Fortune 500 implementation can be built with agentic workflows and run 10 times faster than the first.