Superblocks CEO: Uncover Unicorn Ideas with AI Prompt Analysis
▼ Summary
– Brad Menezes, CEO of Superblocks, believes system prompts used by AI startups hold billion-dollar startup ideas, as they reveal how companies tailor foundational models for specific tasks.
– System prompts, though not always public, are lengthy instructions (5,000-6,000 words) that guide AI models, and Superblocks shared 19 prompts from popular AI coding tools like Windsurf and Cursor.
– Menezes explains that system prompts are only 20% of a startup’s “secret sauce,” with the remaining 80% being “prompt enrichment” infrastructure around LLM calls.
– Studying system prompts involves analyzing role prompting (consistency/personality), contextual prompting (guardrails/task clarity), and tool use (enabling agentic tasks beyond text generation).
– Superblocks targets non-programmers by handling security and enterprise data access, with notable customers like Instacart and internal tools built by business teams using their AI agent.
Brad Menezes, CEO of Superblocks, has uncovered an unconventional goldmine for spotting the next wave of billion-dollar startup ideas—system prompts used by leading AI companies. These detailed instructions, often spanning thousands of words, shape how foundational AI models operate within specialized applications. While not always public, these prompts reveal the unique approaches companies take to tailor AI for specific tasks.
Superblocks recently made headlines by sharing a collection of 19 system prompts from top AI coding tools, sparking widespread interest across the tech community. The move coincided with their $23 million Series A funding round, bringing their total raised to $60 million. Their flagship product, Clark, is an enterprise AI agent designed to help non-developers create applications—a vision inspired by analyzing these very prompts.
Menezes breaks down the anatomy of system prompts into three critical components: role prompting, contextual prompting, and tool use. Role prompting defines the AI’s identity and expertise, like Devin’s prompt positioning it as a “real code-wiz.” Contextual prompting sets boundaries—Cursor’s instructions, for example, emphasize efficiency by avoiding unnecessary code displays. Tool use expands functionality, enabling actions like database queries or shell commands, as seen in Replit’s lengthy prompt.
What surprised Menezes most was realizing that the system prompt itself accounts for just 20% of the magic. The remaining 80% lies in “prompt enrichment”—the infrastructure around API calls, including validation steps and post-response actions. This insight reshaped Superblocks’ strategy, leading them to focus on simplifying app development for business teams.
Internally, Superblocks practices what it preaches. Engineers are barred from building internal tools; instead, business teams use Clark to create custom agents—like one that mines CRM data for leads or another that balances sales engineer workloads. For Menezes, this approach isn’t just about innovation—it’s about empowering non-technical users to solve problems without relying on developers.
While Superblocks hasn’t yet reached unicorn status, its growing client roster, including Instacart and Paypaya Global, suggests the strategy is resonating. By decoding the hidden layers of AI prompts, Menezes believes startups can uncover untapped opportunities—proving that sometimes, the best ideas are hiding in plain sight.
(Source: TechCrunch)
