Brad Menezes, CEO of Superblocks, recently highlighted the underutilized promise of system prompts in AI startups. He challenges us to imagine that the next wave of billion-dollar ideas are hiding beneath these prescriptive structures. His wisdom can’t come soon enough, with the rapid pace of AI development redefining the landscape almost daily. Suspension of work to make foundational models from OpenAI, Anthropic and others prerequisite to success for application-level products in AI.
These system prompts are crucial directives for these opaque AI engines. They instruct the models to produce answers in accordance with certain personas or scenarios. Menezes identified three essential components of system prompts that warrant further study: role prompting, contextual prompting, and tool use. These features are important for an overall picture of how AI is interacting with users and carrying out tasks appropriately.
Menezes was adamant that the system prompts are in plain English. They are incredibly expansive and detailed often exceeding 5,000 to 6,000 words. This degree of detail is important for training AI models properly. For instance, a system-level prompt might tell an AI to role-play as a software engineer. Second, it will use a real computer operating system, not some toy environment. Cursor’s system prompt tells the model to only call tools when it absolutely needs to. Further, it cautions against directly addressing the user in terms of the tool names in the first person.
It’s no wonder Menezes’ tweet has gone viral, tapping into a wave of excitement and interest on this front. Since then, it has racked up close to 2 million views and garnered the interest of some big fish in Silicon Valley. This engagement serves to highlight the importance of examining existing unicorn AI startups to draw conclusions about which prompting strategies produce the best results. By reverse engineering other people’s system prompts, Menezes has had the privilege of learning what works, what doesn’t, and what kind of information excites developers and users both.
At Superblocks, Menezes isn’t just an observer—he’s the one implementing these concepts in practice at his own company. The firm has raised a total of $60 million, including a recent $23 million Series A funding round, to enhance its vibe coding tools aimed at empowering non-developers in enterprise environments. This last development fits perfectly with his view that system prompts can really empower non-programmers to create applications in a powerful way.
Furthermore, Menezes suggests that when constructing system prompts, developers should read relevant file content prior to making edits and tackle any evident errors without over-complicating fixes. He discourages you from winging it, or trying to figure out more than three iterative solutions in a row, and encourages clarity and precision when composing your prompts.
Menezes stresses that three main ingredients are essential for developing system prompts that work well. He’s convinced that 20% of the rest goes to role prompting, contextual prompting, and tool use. The other 80% is what he refers to as “prompt enrichment.” That includes painting a wider picture with the prompts to increase their impact.
Superblocks makes it easy to create mobile apps with little or no programming skills. It provides many capabilities to address complicated requirements, including enterprise security and access to other enterprise data sources, including Salesforce. This mission underscores the company’s devotion to democratizing access to technology and enabling innovation from non-technical users.
The AI field is indeed exponentially growing. Founders at startups who are interested in pursuing AI for the better will have to learn the ins and outs of system prompts sooner rather than later. Menezes’ insights could guide entrepreneurs in developing innovative solutions that resonate with users while navigating the complex landscape of AI development.
