Superblocks CEO: Discovering Unicorn Ideas Through AI System Prompts

Brad Menezes, the CEO of Superblocks, an enterprise vibe coding startup, is convinced that the next wave of billion-dollar startup ideas is hidden in plain sight: within the system prompts of existing unicorn AI startups.

System prompts are extensive instructions, often exceeding 5,000-6,000 words, used by AI startups to guide foundational models from companies like OpenAI or Anthropic in generating application-level AI products. Menezes views these prompts as a master class in prompt engineering.

"Every single company has a completely different system prompt for the same foundational model," Menezes explained to TechCrunch. "They’re trying to get the model to do exactly what’s required for a specific domain, specific tasks."

While system prompts are not entirely hidden, they are not always publicly available. However, customers can request them from many AI tools. In a strategic move, Superblocks shared a file of 19 system prompts from popular AI coding products like Windsurf, Manus, Cursor, Lovable, and Bolt as part of their new product announcement for an enterprise coding AI agent named Clark.

Menezes's tweet about this initiative went viral, garnering nearly 2 million views and catching the attention of notable figures in Silicon Valley. Recently, Superblocks announced a $23 million Series A extension round, bringing their total Series A funding to $60 million, aimed at enhancing their vibe coding tools for non-developers in enterprises.

In an interview, Menezes shared insights on studying system prompts to derive valuable lessons. He emphasized that the system prompt itself constitutes only about 20% of the "secret sauce." The remaining 80% involves "prompt enrichment," which includes the infrastructure built around the calls to the LLM, such as user prompt instructions and response actions like accuracy checks.

Menezes highlighted three key components of system prompts to focus on: role prompting, contextual prompting, and tool use. Role prompting ensures consistency by giving the LLMs a purpose and personality. For example, Devin’s prompt begins with, "You are Devin, a software engineer using a real computer operating system. You are a real code-wiz: few programmers are as talented as you at understanding codebases, writing functional and clean code, and iterating on your changes until they are correct."

Contextual prompting provides models with the necessary context before acting, offering guardrails to reduce costs and ensure task clarity. For instance, Cursor’s prompt instructs, "Only call tools when needed, and never mention tool names to the user — just describe what you’re doing. … don’t show code unless asked. … Read relevant file content before editing and fix clear errors, but don’t guess or loop fixes more than three times."

Tool use enables agentic tasks by instructing models on how to go beyond text generation. Replit’s prompt, for example, details editing and searching code, installing languages, setting up and querying PostgreSQL databases, executing shell commands, and more.

By studying others’ system prompts, Menezes observed that tools like Loveable, V0, and Bolt emphasize fast iteration, while Manus, Devin, OpenAI Codex, and Replit focus on helping users create full-stack applications, though the output remains raw code.

Menezes sees an opportunity for non-programmers to write apps if his startup can manage more complex tasks, such as security and access to enterprise data sources like Salesforce. While Superblocks is not yet a multi-billion-dollar startup, it has secured notable clients, including Instacart and Paypaya Global.

Internally, Superblocks is also using its product. Engineers are restricted from writing internal tools, focusing instead on building the product. Business teams have developed agents for various needs, such as using CRM data to identify leads, tracking support metrics, and balancing sales engineer assignments.

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

_Correction: This story was updated to clarify that the most recently announced round was an extension round and to update the Series A total amount raised._