"What is Anthropic?" This seemingly innocuous question, posed by an executive at a billion-dollar company, encapsulates the profound disconnect Brendan Falk discovered between Silicon Valley's AI fervor and the ground truth of enterprise adoption. Falk, founder and CEO of Zeus (now Hercules.app), shared his candid journey from leading AI initiatives at Amazon's private equity arm to attempting to build the "Palantir for AI" for Global 2000 companies, ultimately leading to a significant pivot. His conversation with Alessio Fanelli and Swyx of Latent Space revealed the often-unseen complexities and economic pitfalls of large-scale AI transformations.
Falk's initial vision for Zeus was compelling: target Fortune 500 companies with $5-10 million contracts, build custom AI agents using a proprietary platform, and maintain high margins. His background, including the acquisition of his previous company Fig by AWS and a role as Global AI Lead at Amazon's private equity business development team, provided him with unparalleled insight into the diverse needs of hundreds of companies seeking AI integration. He saw a clear market opportunity, assuming that if AI could deliver immense value, large enterprises would readily embrace bespoke solutions.
However, the reality of enterprise AI proved far more challenging than anticipated. One core insight Falk highlighted was the sheer lack of fundamental AI understanding in many large, established businesses. While tech insiders are immersed in the latest models and capabilities, many executives in traditional industries are still grappling with the basics, as evidenced by the "What is Anthropic?" query. This foundational gap means that even initiating an AI transformation requires extensive education and hand-holding, significantly lengthening sales cycles and increasing upfront effort.
Another critical learning was the role of traditional system integrators (SIs) in stifling genuine AI progress. Falk observed that many large enterprises, lacking in-house technical expertise, defer to SIs like Accenture. These SIs, incentivized by billable hours, often propose numerous Proofs of Concept (POCs) that are essentially "science experiments" and rarely make it to production. "The board was yelling at the CEO saying where's our AI strategy," Falk recounted, describing a cascading pressure that often resulted in SIs "bleeding these companies dry" with costly, non-scalable solutions. The fundamental problem was a misalignment of incentives: SIs profit from extended engagements, not rapid, efficient deployment.
The "land and expand" strategy, common in SaaS, also proved ineffective for Zeus's initial model. Each enterprise use case, from insurance RFPs embedded in Word documents to complex legal contract reviews, was unique. This meant that every new deployment required custom development, negating the benefits of a proprietary platform designed for scalability. "The amount of work for a $500K contract versus a $10M contract isn't proportional - it's almost the same," Falk stated, underscoring the unsustainable economics of highly customized, low-repeatability projects. The hidden complexity of messy data, company-specific edge cases, and the inability of AI to magically solve existing integration problems meant that full-time product managers were needed for each client, further eroding margins.