Superblocks CEO: How to find a unicorn idea by studying AI system prompts

Brad Menezes, the CEO of Superblocks, a startup focused on enterprise coding, believes that the next wave of billion-dollar business concepts might be more accessible than one would think. He points to the system prompts utilized by successful AI startups as a treasure trove of inspiration. These prompts, which often exceed 5,000 to 6,000 words, guide foundational models from prominent companies like OpenAI and Anthropic in crafting their AI-driven applications. Menezes likens these system prompts to a master class in prompt engineering. "Every company utilizes a unique system prompt for the same foundational model," he explained in a recent interview with TechCrunch. These tailored prompts are designed to elicit specific responses for distinct domains and tasks. While many AI tools allow customers to request access to their system prompts, not all of them are readily available to the public. In a move to share knowledge, Superblocks recently released a file containing 19 system prompts from popular AI coding tools, such as Windsurf and Manus. This initiative caught significant attention online, with Menezes's tweet garnering nearly 2 million views and interactions from influential figures in Silicon Valley, including Sam Blond and Aaron Levie. Superblocks also announced a successful $23 million Series A funding round last week, bringing its total capital to $60 million. This funding is aimed at enhancing its vibe coding tools, which are designed for non-developers within enterprises. When asked about the insights gained from analyzing other companies' system prompts, Menezes mentioned a crucial takeaway: the system prompt itself constitutes only about 20% of the overall effectiveness. According to him, the remaining 80% relates to what he terms "prompt enrichment." This encompasses the infrastructure a startup builds around interactions with the language model, including user instructions and response accuracy checks. He highlighted three key components to focus on when studying system prompts: role prompting, contextual prompting, and tool use. Menezes emphasized that while system prompts are articulated in natural language, they require a high degree of specificity. "You have to communicate as if you're speaking to a human colleague, and the instructions must be flawless," he stated. Role prompting establishes consistency for the language models by giving them defined purposes and personalities. For instance, one prompt introduces a character as a highly skilled software engineer tasked with writing and refining code. Contextual prompting equips models with the necessary background to make informed decisions. This aspect can help minimize costs and clarify tasks. For example, one prompt advises users to call tools only when necessary and to avoid mentioning tool names, focusing instead on describing actions. The final component, tool use, allows for more autonomous tasks by directing the models on how to execute beyond mere text generation. By analyzing the system prompts of other coding tools, Menezes identified varying emphases across products. For instance, while some tools prioritize rapid iteration, others focus on assisting users in building comprehensive applications, albeit with raw output. Though Superblocks has not yet reached the ambitious heights of a multi-billion dollar venture, it has secured notable clients such as Instacart and Paypaya Global. Interestingly, Menezes has adopted a unique internal approach, forbidding his engineers from creating internal tools. Instead, the business team has developed agents to cater to their operational needs, including one for identifying leads from CRM data and another for tracking support metrics. "This allows us to create our own tools rather than purchasing them," he concluded.