Brad Menezes, CEO of Superblocks, isn’t chasing the next unicorn idea—he’s dissecting it. In an industry obsessed with big language models, it’s not the AI itself that captivates Menezes, but rather the instructions shaping its behavior. Known as system prompts, these behind-the-scenes scripts are often the secret fuel behind AI’s most impressive feats. And Menezes believes they hold the keys to the next wave of billion-dollar startups.
System prompts are no ordinary snippets of code. They’re extensive—sometimes over 6,000 words—and written in plain language that mimics conversations with a capable colleague. At their core, they teach large language models (LLMs) how to behave, think, and act within specific domains. From customer service to coding, these prompts define the personality and intelligence of today’s top AI agents.
But here’s the kicker: they’re not locked in a vault.
“Every single company has a completely different system prompt for the same foundational model”
Menezes told Voke Magazine. That means the real innovation isn’t just in the tech—it’s in the tuning. And by reviewing these prompts, founders can reverse-engineer the DNA of high-performing AI startups.
To prove his point, Menezes recently released 19 system prompts from AI coding agents like Windsurf, Cursor, and Bolt alongside the launch of Superblocks’ own agent, Clark. The release triggered a viral storm, with 2 million views and attention from high-profile names in tech like Sam Blond and Aaron Levie. For a startup still defining its identity, the signal was clear: the market is hungry for transparency—and tactical insight.
What Menezes discovered by sifting through these prompts wasn’t just a competitive advantage; it was a masterclass in product design. Only about 20% of the success lies in the system prompt itself, he found. The other 80%? It’s what he calls “prompt enrichment”—a blend of middleware infrastructure, contextual layering, and accuracy protocols that elevate the AI from functional to phenomenal.
Three elements stood out: role prompting, contextual prompting, and tool orchestration.
Role prompting assigns the AI a professional identity—like Cursor’s instruction to act as a disciplined engineer who won’t show unnecessary code. Contextual prompting, on the other hand, provides the “why” and “when,” ensuring the AI stays grounded and relevant. Then there’s tool use: telling the model how to interact with databases, codebases, and third-party apps to complete complex workflows.
These aren’t just academic insights. Superblocks is using them to shift the paradigm for enterprise software. Their engineers no longer write internal tools—business users do, powered by Clark. Internal agents built on top of CRM data now help qualify leads, track support tickets, and manage workflows with zero developer hours.
It’s a radical flip of the traditional enterprise software model. As Menezes puts it,
“This is basically a way for us to build the tools and not buy the tools”
For a company that just extended its Series A to $60 million, with backers like Kleiner Perkins and connections to AI elite, Superblocks is turning system prompt study into a science—and a strategy. While the next unicorn might not be visible on the surface, Menezes makes one thing clear: in the prompt lies the blueprint.
