Brad Menezes, the CEO of enterprise vibe coding startup Superblocks, believes the next great wave of billion-dollar startup ideas may be hiding in plain sight—embedded within the system prompts used by rapidly growing AI companies.
System prompts, typically complex instructional texts extending to thousands of words, serve as the foundational guidelines that AI startups use to direct artificial intelligence models from providers such as OpenAI and Anthropic. These prompts aren’t a secret per se; customers can request these instructions from many AI systems. However, they are rarely shared openly.
In conjunction with unveiling its latest AI product, an enterprise coding agent named Clark, Superblocks released a valuable collection of system prompts from well-known AI ventures including Windsurf, Manus, Cursor, Lovable, and Bolt. When Menezes shared these insights online, the response was extraordinary, drawing attention from nearly two million viewers, including established Silicon Valley leaders and investors.
Having recently secured $23 million in Series A funding (part of a total $60 million investment), Superblocks is gaining momentum by making advanced coding tools accessible to non-developers within enterprises. Explaining the significance of studying these system prompts, Menezes said they provide crucial clues for refining how AI products perform specific tasks. However, he indicated the prompts are only roughly 20% of the equation—the remainder lies in what he termed “prompt enrichment,” or infrastructure built around the system prompts, including additional instructions, accuracy checks, and adjustments based on user interactions.
According to Menezes, the analysis of system prompts can be approached through three components—role prompting, contextual prompting, and tool utilization.
Role prompting provides the AI models with a consistent “identity” and clear function. For example, Devin’s prompt explicitly describes the AI persona as a skilled software engineer, setting expectations for the quality of responses the model should deliver.
Contextual prompting equips models with essential background and constraints for decision-making, reducing ambiguity and operational costs. Cursor’s prompt, for instance, explicitly instructs the AI on careful resource usage, efficient tool calls, and interaction transparency with the user.
Lastly, tool utilization defines actionable tasks that go beyond simply generating textual responses. Replit’s prompt extensively details capabilities such as file editing, software installation, database management, and command execution, ensuring the AI can perform complex operations.
By studying how various companies structured their system prompts, Menezes quickly identified strategic patterns. Certain tools, like Loveable and Bolt, emphasized rapid iteration, while others like Manus and Replit offered users the raw flexibility to construct comprehensive applications. Recognizing an untapped opportunity here, Menezes said Superblocks could make app development accessible to non-programmers by leveraging these insights. Beyond merely enabling AI-driven app-generation, his platform aims to integrate critical enterprise-grade features, such as robust security measures and compliance with corporate data systems like Salesforce.
Superblocks, though not yet among the ranks of billion-dollar startups itself, already serves prominent customers, including brands such as Instacart and Paypaya Global. Additionally, the company extensively employs its own platform internally. For instance, Menezes has mandated that his technical staff focus strictly on external products, leaving internal tooling and business systems to be completely automated by Superblocks agents—everything from lead generation tools using customer-relations management systems to internal support and workload-balancing applications.
“We build rather than buy our internal tools,” Menezes explained, highlighting the power of his own approach to AI-driven app creation.
