Unlocking Unicorn Startup Ideas Through AI System Prompts

Superblocks CEO Brad Menezes believes the next wave of billion-dollar startup ideas lies within AI system prompts. These detailed prompts, often exceeding 5,000 words, instruct foundational AI models like those from OpenAI and Anthropic. Menezes sees them as a masterclass in prompt engineering.

These prompts, used by existing unicorn AI startups, reveal how companies tailor foundational models for specific applications and tasks. While some AI tools share their prompts upon request, they aren't always publicly accessible.

To shed light on this area, Superblocks released a collection of 19 system prompts from popular AI coding products like Windsurf, Manus, Cursor, Lovable, and Bolt, alongside their announcement of Clark, an enterprise AI coding agent. This move, shared by Menezes on X (formerly Twitter), garnered significant attention, including from prominent figures in Silicon Valley.

Deconstructing System Prompts for Insights

Menezes emphasizes that the system prompt itself represents only about 20% of the "secret sauce." The remaining 80% lies in "prompt enrichment," the infrastructure built around interacting with the large language model (LLM). This includes instructions added to user prompts and actions taken upon receiving the LLM's response, such as verifying accuracy.

He highlights three key aspects of system prompts to analyze:

  • Role prompting: Defines the LLM's purpose and personality, ensuring consistent behavior.
  • Contextual prompting: Provides context and guardrails, influencing factors like cost and task clarity.
  • Tool use: Instructs the model on how to utilize external tools, extending its capabilities beyond text generation.

System prompts, while written in natural language, require exceptional specificity. Menezes explains, "You have to speak as if you would to a human co-worker. The instructions have to be perfect."

By studying other companies' prompts, Menezes identified a focus on fast iteration in tools like Loveable, V0, and Bolt, while Manus, Devin, OpenAI Codex, and Replit prioritize full-stack application creation, albeit with raw code output. This revealed an opportunity to empower non-programmers to build apps by addressing additional needs like security and enterprise data integration.

Superblocks, while not yet a multi-billion dollar venture, boasts notable clients like Instacart and Paypaya Global. Internally, the company uses its own product extensively, with business teams building AI agents for various tasks, from lead identification and support metric tracking to sales engineer assignment balancing.

Menezes concludes, "This is basically a way for us to build the tools and not buy the tools."