'''# Why Your Approach to Prompt Engineering Is Killing Your Potential
Hot take: Your prompt engineering strategy is likely flawed. It's time to address the silent killers of creative output and discover what the top 1% are doing differently.
I’ve seen it more times than I can count. A founder, brilliant and driven, is trying to get their AI to generate a logo, an ad creative, or even just a decent blog post. They’re wrestling with the prompt, tweaking a word here, a phrase there, and getting… garbage. The frustration is palpable. They’re convinced the AI is dumb. But the hard truth? The problem isn’t the AI. It’s the approach.
I remember back in the early days of MovieLaLa, we were trying to generate movie posters that would resonate with our audience. We were using a very early version of what would now be considered a primitive image generation model. Our prompts were simple, something like "a poster for a romantic comedy". The results were laughably bad. Cheesy, generic, and completely off-brand. We almost gave up on the idea entirely.
It wasn't until we started thinking differently about how we "talked" to the AI that things started to change. We realized that prompting wasn't just about telling the AI what to do. It was about teaching it, guiding it, and giving it the right context. It was a conversation, not a command.
That shift in mindset was a game-changer for us. And it’s a lesson that I see so many founders today still haven’t learned. They’re stuck in the old way of thinking, and it’s costing them dearly.
The Old Way of Prompting: A Recipe for Mediocrity
So what does this "old way" of prompting look like? It’s probably familiar to you. It’s the one-shot prompt, the laundry list of keywords, the vague and ambiguous instruction. It’s the kind of prompt that looks like this:
- "A logo for a tech startup, modern and innovative"
- "An ad for a new coffee shop, cozy and inviting"
- "A blog post about the future of AI, optimistic and insightful"
These prompts are the equivalent of asking a designer to "make something cool". You’re leaving everything up to interpretation. You’re not providing any real direction. And you’re setting yourself up for disappointment.
Why is this approach so ineffective? Because it ignores the fundamental nature of how these models work. They’re not mind readers. They’re pattern matchers. They’ve been trained on vast amounts of data, and they’re looking for patterns in your prompt that they can connect to the patterns they’ve seen before.
When you give them a vague prompt, you’re not giving them enough information to find the right patterns. You’re forcing them to guess. And when they guess, they’re going to default to the most common, the most generic, the most… mediocre.
I’ve seen this firsthand with some of the companies I’ve invested in. A founder will come to me, excited about their new AI-powered design tool. They’ll show me a gallery of generated images, and they’ll all have that same generic, AI-generated feel. The spark of originality is missing. And it’s because they’re not pushing the models hard enough. They’re not giving them the right inputs to create something truly unique. ''' '''
The New Way: Prompting as a System
So if the old way is broken, what’s the alternative? It’s about moving from one-shot prompts to a more systematic approach. It’s about treating prompting not as a single action, but as a process of continuous refinement and iteration. It’s about building a “prompting system.”
What does this system look like? It has three key components:
- Deep Context: Giving the AI a rich, detailed understanding of what you’re trying to achieve.
- Hard Constraints: Setting clear boundaries and rules to guide the AI’s creative process.
- Iterative Refinement: Continuously tweaking and improving your prompts based on the AI’s output.
Let’s break these down.
Deep Context: The Unfair Advantage
Most people give the AI surface-level context. They’ll say “a logo for a tech startup”. The top 1% give the AI deep context. They’ll say something like:
“We are a B2B SaaS startup called ‘Momentum’ that provides real-time analytics for sales teams. Our brand personality is professional, trustworthy, and slightly futuristic. We want a logo that is minimalist, abstract, and uses a color palette of dark blue (#001F3F), silver (#DDDDDD), and a bright accent color like orange (#FF851B). The logo should not contain any literal representations of charts or graphs. It should feel like it belongs in the same family as logos from companies like Stripe and Scale AI.”
See the difference? The second prompt is dripping with context. It’s giving the AI a deep understanding of the business, the brand, and the desired aesthetic. It’s not just telling the AI what to do, it’s telling it why.
When I was advising an early-stage company in the AI video generation space, we spent a whole week developing a “context document” before we even wrote a single prompt. This document included everything from our target audience’s psychographics to the emotional arc we wanted our videos to follow. It was our “bible” for all things AI-related. And it made all the difference. Our generated videos went from generic and soulless to deeply engaging and on-brand.
Hard Constraints: The Power of “Don’t”
One of the most counterintuitive but powerful prompting techniques is to tell the AI what not to do. By setting hard constraints, you’re forcing the AI to think more creatively. You’re pushing it out of its comfort zone and into new, unexplored territory.
For example, instead of just saying “a picture of a futuristic city”, you could say:
“A picture of a futuristic city, but with no flying cars, no neon signs, and no skyscrapers. The architecture should be inspired by ancient Roman and Babylonian designs. The mood should be serene and utopian, not dystopian.”
By adding these negative constraints, you’re forcing the AI to come up with a much more original and interesting interpretation of the prompt. You’re guiding its imagination, not just letting it run wild.
I’ve used this technique to generate some of the most unique and compelling images I’ve ever seen. It’s a simple trick, but it’s incredibly effective. It’s the difference between getting a generic sci-fi cityscape and getting a piece of art that makes you stop and think.
Iterative Refinement: The Feedback Loop of Excellence
The final piece of the puzzle is iterative refinement. You’re not going to get the perfect output on your first try. You need to be prepared to experiment, to learn from the AI’s mistakes, and to continuously refine your prompts.
This means treating the AI not as a vending machine, but as a creative partner. It means having a conversation with it. It means looking at its output, figuring out what’s working and what’s not, and then adjusting your prompt accordingly.
For example, if you’re trying to generate a blog post and the AI is using too much jargon, you can add a constraint like “Explain this concept in simple terms that a 10th grader could understand.” If the tone is too formal, you can say “Write in a more casual, conversational style.”
This feedback loop is where the magic happens. It’s where you go from being a passive user of AI to an active collaborator. It’s where you start to unlock the true potential of these models. ''' '''
Case Study: From Generic to Genius at a Portfolio Company
Let me give you a real-world example. I’m an investor in a company—let’s call them “Canvas”—that’s building an AI-powered tool for creating ad creatives. When they first pitched me, their technology was impressive, but their output was… forgettable. The ads they generated were the definition of generic. They looked like stock photos with some text slapped on top.
I invested, but with a condition: they had to let me come in and work with their team on their prompting strategy. The founder, a brilliant engineer, was skeptical. He thought the problem was with the model, not the prompts. But he agreed.
We spent the first two days not writing a single prompt. Instead, we built a “Brand Bible” for a fictional company. We created a detailed persona for their target customer. We defined their brand voice, their color palette, their typography. We even wrote a list of “brand words” that they should and shouldn’t use.
Then, we started prompting. But instead of just saying “an ad for a new energy drink”, we would feed the AI the entire Brand Bible as context. Our prompts looked more like this:
Prompt:
Context: [Insert entire Brand Bible here]
Task: Generate an ad creative for our new energy drink, “Flow”. The ad should feature a young woman in her late 20s, a freelance graphic designer, working in her minimalist home office. She should look focused and energized, but not hyper. The mood should be calm and productive. The color palette should be dominated by our primary brand colors: #F5F5F5 (off-white) and #4A4A4A (dark gray), with our accent color, #00BFA5 (teal), used sparingly. The ad should include the headline “Find Your Flow” in our brand font, “Montserrat”.
Negative Constraints:
- No images of people partying or doing extreme sports.
- No bright, flashy colors.
- No generic “power up” or “get energized” slogans.
The difference was night and day. The AI went from generating cheesy, generic ads to creating beautiful, on-brand creatives that looked like they had been designed by a top-tier agency. The founder was blown away. He finally understood that prompting wasn’t just about telling the AI what to do. It was about teaching it how to think.
Canvas went on to raise a massive Series A round, and their tool is now being used by some of the biggest brands in the world. And it all started with a shift in their approach to prompt engineering.
Your Action Plan: Stop Tinkering, Start Building
So how can you apply this to your own business? It’s time to stop tinkering with one-off prompts and start building a systematic approach to prompt engineering. Here’s your action plan:
- Create a “Brand Bible” for your AI. This should be a living document that contains everything the AI needs to know about your brand, your audience, and your goals. It should be the single source of truth for all of your AI-powered content creation.
- Embrace negative constraints. Don’t be afraid to tell the AI what not to do. This is one of the most powerful ways to guide its creativity and generate more original outputs.
- Build a feedback loop. Treat the AI as a creative partner. Have a conversation with it. Continuously refine your prompts based on its output. This is how you’ll go from being a passive user to an active collaborator.
This isn’t about finding the “perfect” prompt. It’s about building a system that allows you to consistently generate high-quality, on-brand content at scale. It’s about moving from a world of one-shot prompts to a world of prompt systems.
I’ve seen this approach transform businesses. I’ve seen it unlock new levels of creativity and innovation. And I know it can do the same for you.
The next time you’re staring at a blank prompt, don’t just ask yourself “What should I tell the AI to do?” Ask yourself, “What does the AI need to know to do its best work?”
That’s the question that separates the amateurs from the pros. That’s the question that will unlock your true potential. '''
Frequently Asked Questions
How has this view evolved over time?
My thinking on most topics has changed significantly over the years. Early in my career, I held many conventional views that experience proved wrong. I try to update my beliefs when the evidence changes.
What's the most common pushback you get on this?
People often push back by citing exceptions or edge cases. And they're usually right that exceptions exist. But building a strategy around exceptions rather than patterns is a losing game for most founders.
Do all experts agree with this view?
No, and that's fine. The best ideas in business are often contrarian. I share my perspective based on my experience and data, but I encourage you to seek out opposing viewpoints and form your own conclusions.