The Hard Truths of Enterprise AI: Why "Palantir for AI" Failed

Brendan Falk, founder and CEO of Zeus (now Hercules), shared a candid and illuminating account of his pivot away from an ambitious venture to build an "AI-native Palantir" for Global 2000 companies. In an interview with Alessio Fanelli of Kernel Labs and Swyx of Smol AI, Falk detailed the unexpected complexities and economic realities that led him to abandon his initial vision after just six months. His journey, marked by both a successful acquisition by AWS with his previous company Fig and this subsequent pivot, offers invaluable lessons for founders, investors, and AI professionals navigating the rapidly evolving enterprise AI landscape.

Falk's initial foray into enterprise AI was informed by a unique vantage point: his role as Global AI Lead for Amazon's Private Equity Business Development team. This position granted him unparalleled access to "tens, probably hundreds of executives" across "tens of companies" – about 20,000 private equity-backed AWS customers. His deliberate intent was to immerse himself in diverse industries, geographies, and company sizes to identify the next big opportunity. He observed that these PE-backed firms, driven by the imperative to maximize returns for eventual sale, often moved with a speed uncharacteristic of traditional large enterprises.

However, this broad exposure quickly unveiled a stark reality. Many companies, despite revenues exceeding $1 billion, operated with a profound lack of AI literacy. Falk recounted instances where executives would genuinely inquire, "Yes, so what is Anthropic?" This fundamental gap in understanding meant that while boards demanded an AI strategy, the operational teams often lacked the foundational knowledge or internal technical capabilities to execute. These non-technical enterprises frequently turned to system integrators like Accenture, who, as Falk noted, were effectively "bleeding these companies dry" with numerous Proof-of-Concepts (POCs) that rarely, if ever, reached production. The traditional system integrator model, he explained, is incentivized to "sell more headcount for longer," not to deliver high-quality, production-ready solutions.

The inherent complexity of enterprise processes further compounded these issues. Falk highlighted the "messy data problem," where databases were riddled with missing, conflicting, or incorrect information. Beyond data, companies faced myriad "edge cases that are incredibly difficult to account for up front." He vividly described an insurance RFP where "an Excel spreadsheet was embedded in the Word document," a technical absurdity he "didn't even know existed." This intricate web of company-specific peculiarities meant that AI solutions, while theoretically powerful, demanded extensive custom development and ongoing human intervention.

This bespoke development model clashed directly with the economic realities of scaling a services business. Falk discovered that "the amount of work for a $500K contract versus a $10M contract isn't proportional – it's almost the same." The effort involved in understanding complex business processes, navigating organizational politics, cleaning messy data, building custom agents, and managing change was largely fixed, regardless of contract size. This meant that pursuing smaller contracts, while seemingly an easier entry point, was simply not lucrative enough to sustain the business. Conversely, securing the truly large, multi-million dollar contracts required a level of deep specialization and brand recognition that his nascent company, Zeus, couldn't immediately command. The vision of becoming an "AI-native Palantir" for any use case, across any industry, became economically unsustainable due to the sheer lack of repeatability.

Falk also recognized the "existential threat" posed by hyper-specialized vendors. He realized that for many specific AI use cases, companies like Vapi, Decagon, or Sierra, with "hundreds of millions focused on one thing," would always outperform a generalist AI services provider. His team was essentially building bespoke versions of what these focused companies offered, but without their concentrated resources or deep domain expertise. This insight underscored the difficulty of competing across a broad front in the rapidly evolving AI landscape.

Ultimately, these hard-won lessons led Falk to pivot Zeus into Hercules.app, an AI-native website builder. This move marked a return to his roots in developer tools, a space where his previous company, Fig, had found significant success. Hercules aims to democratize website creation, allowing users to "one-shot" build fully functional, production-ready websites simply by chatting with an AI. By white-labeling backend infrastructure like Convex.dev, Hercules streamlines the technical complexities, focusing on the user experience. This pivot targets the vast market of SMBs and individuals who need a website but lack technical skills or the budget for traditional development. In this new venture, the challenges of messy enterprise data, endless custom integration, and the system integrator trap are replaced by a focus on intuitive product design and high gross margins, a model where the brand and ease of use can truly differentiate.