Stable Discussion

Stable Discussion

Fast Prototypes that Start Conversations

How I pulled context from my task assistant, established prototype descriptions, and quickly iterated with Magic Patterns to create incredible web pages to get conversations started on my next startup

Stable Discussion's avatar
Ben Hofferber's avatar
Stable Discussion
and
Ben Hofferber
Sep 03, 2025
∙ Paid
2
1
Share

I’ve been diving into Learn More Faster this week as I have a few startup ideas that I want to validate. The book covers the “Bullseye Customer Sprint” methodology for helping early-stage startups effectively identify ideal customers and how they can offer those customers value. An essential part of building any business.

While the 100 page PDF is free to read, I also enjoyed listening along to the NotebookLM podcast narration on the same topic which you’ll find at the end of this post. LLMs can help us to break into new concepts clearly and it feels a little bit like the process described in How to Read a Book. When you first pick up a new book they suggest doing a scan of the table of contents, index, and summary of the book to get your bearings. We can now take this one step further!

Let me walk you through how I built a prototype using this methodology—and show you just how powerful AI made the process.

Step 1 - Defining my Product

One idea that’s stayed with me is making it easier for engineering leaders to find the right candidates. In some organizations these leaders don’t get much help and need to figure out how to recruit great people on their own. Suddenly the first 30-60 minutes of their day is scrolling through LinkedIn messages, job boards, or massive piles of resumes. It’s a far cry from the cool tech stuff they signed up for but it’s also an essential part of building their team.

In my prior role as Head of Engineering, I developed a proof of concept for this product, but I realized I hadn’t truly validated the idea with my target audience. But, I had a working example and a lot of good source material on my product idea.

I had started with a specific candidate in mind (me) and had noted key factors to consider. Throughout the prototyping process, I used AI-assisted coding and documented both how the application works and why it might appeal in a crowded market of hiring agencies and recruiters. These notes were essential for AI coding and could now also be leveraged to help me define the product again here.

I relied on Granola to turn conversations about the idea into actionable notes. Talking through ideas helped me refine my approach and capture valuable insights.

After gathering these notes, my next step was to create a clear product description using my task management assistant, Benny Chat.

If you’re wondering why my AI is a little mean, I’ve tweaked the chat personality just a bit to regularly roast me a bit. It’s been a fun addition to my daily task management!

I asked it to pull up relevant notes and it quickly brought up some likely candidates from its search over all of my notes. Next I needed a summary to ensure these notes were the ones I needed.

After some back and forth, I found the note I was looking for but also got a bit of helpful related context along the way from other research I’ve done.

Benny offered to write me a pitch for this business and that’s exactly what I was looking to do. I adjusted it slightly to focus on the niche I’m aiming at and let it get me the following definition:

AI Agency for Engineering Leaders: Business Definition A talent-matching agency for engineering leaders, powered by AI and human expertise, that delivers highly targeted shortlists of technical candidates explicitly tailored to the unique skill criteria and team culture requirements set by the hiring manager.

  • Engineering leaders describe their specific hiring criteria (skills, experience, values, context).

  • Each leader is paired with an AI agent—and backed by human experts—that actively sources, filters, and ranks a weekly curated slate of “best-fit” candidates (no noisy spam, just right-fits).

  • Recommendations are grounded in data but filtered and contextualized to what really matters in the role.

  • The service goes beyond “keyword matching”—it helps managers clarify their actual needs, guides tradeoff decisions, and offers transparent reasoning for every candidate picked or skipped.

We have a Product Definition and now we’re ready for the next step!

As an aside, if this idea could benefit you or you’d like to hear more please feel free to send me a message or find our contact information on Hint Services!

I’d love to connect!

Step 2 - Defining our Prototypes

Learn More Faster describes how customers will give you better feedback when they have three different ideas to compare and contrast together. A single example doesn’t open up enough conversation, two tends to have users just pick a favorite, so three is the sweet spot where users start to step back and think about pros and cons of each approach. In the real world customers are often comparing multiple products to evaluate which one they should use and this selection of prototypes mimics the natural landscape of their decision making.

Generated with Midjourney

Now, three prototypes can be a lot of work especially if you have one idea that you absolutely love and it’s hard to break out of that option. This is a great opportunity to leverage the thinking of an LLM to help broaden your available options. Specifically a “thinking” AI model that can reason.

In Benny Chat, we offer a Reasoning model that can “think” and leverage longer trains of thought. This allows it to come up with a better answer using the available context.

Currently our Reasoning capability has been performing really well using the new fully released gemini-2.5-pro model from Google but I’m open to changing it when a better model shows up.

Fortunately, this is a widely read book and I’m able to reference the methodology directly as part of my prompt to help the AI coming up with prototypes. Benny gave me some amazing definitions like this one:

I continued brainstorming with our Reasoning model, I came up with a set of three great prototype ideas that could be clearly communicated. When satisfied, my next prompt to Benny Chat’s Reasoning model is:

“Think carefully about the first option and describe for me a landing page that highlights this offering. I'd like you to highlight how the page looks, what is in each section, and define marketing copy that makes sense for this kind of an offering. Think about what we know about Learn More Faster to define a page that is easy to understand and gets the point across clearly to the viewer”

This is going to give us a clear Prototype Definition with a few key details:

  • A overall goal

  • A overall design look, aesthetic, and visual feel

  • Sections by section outline with details of what should be written

  • Actions users will have to engage

Taking the result of this query, we’re ready to get an AI builder involved to get a real prototype generated!

Step 3 - Building the Prototypes

According to Learn More Faster, you don’t need to fully build prototypes—you can simply tweak a competitor’s site, mock up a page in Photoshop, or use a slideshow to simulate the experience. But today, rapid prototyping tools make it easier than ever to quickly create and refine your ideas.

For proficient AI generation, context is everything. In the AI product space, everyone is all-hands-on-deck building Context Engineering into their smart experiences. Fortunately for us, we’ve already done a lot in the previous two steps to provide context that will guide the prompt we’ll use to build out our prototypes!

I start by opening Magic Patterns, dropping in our prototype definition as the prompt, and specifying Shadcn as the design library since it’s my preferred choice for building.

I’ll shortly explain why I’m leveraging Magic Patterns when there are so many other great options available.

Once the prompt generates the webpage, the result is much more robust than our simple outline. The page is organized into clear sections that directly reflect our prototype structure, making it easy to see how each part comes together.

Next, I review the prototype for any obvious inconsistencies or errors. While AI tools are impressive, they can sometimes miss details or generate inaccuracies that need to be corrected. At this stage, I focus on the content rather than design—making sure the statements are accurate and aligned with the overall concept.

Tip: Be specific when making edits—explain what you’re changing and why. The more detail you provide at this stage, the better the model can adapt in future iterations. Thoughtful feedback now will improve results down the line.

Now it’s time to lean into Magic Patterns fully and see why we’re using this specific builder. Using the command /Inspiration we’ll spin up 4 different versions of the page we’ve been building out as our prototype.

Using 4 credits to generate multiple versions is well worth it compared to the time you’d spend forking a solution generating options manually. Magic Patterns makes it easy to explore different approaches and visually communicate ideas that might otherwise stay stuck in your notes.

At this point I keep iterating until I’m satisfied that the prototype meets my expectations. The entire process (not including creating the source notes) can take as little as 15 minutes! And the results speak for themselves!

Transforming Startup Creation

AI doesn’t directly build a startup for you. The skills you need to build a successful startup go far beyond just putting together a good looking web page. It’s about communicating an idea clearly to a specific niche customer and growing out from that point.

AI helps you to find that customer and connect with them faster. Communicate with them better by being able to present more clearly and more directly than you might otherwise be able to. Once you’re in touch with that target customer everything changes, but prototyping is all about making that connection happen to get real insights.

AI tools like Granola can help package that conversation up for you and remind you of details that you would otherwise miss. These tools don’t run the interview for you and they shouldn’t. They won’t tell you if someone is going to actually buy what you’re selling.

I’m excited by the opportunity to connect more effectively with potential customers—leaving a strong impression that I care about their challenges and am committed to solving them. Any tool that helps me do this faster is a winner in my book.

This Substack is reader-supported. To receive new posts and support my work, consider becoming a free or paid subscriber.

If you found this post helpful and want to dive deeper, consider a paid subscription. You’ll get full access to the Stable Discussion Discord—where you can connect with other AI enthusiasts—and to the ideation tool I used today, Benny Chat. With prices going up soon, now is the perfect time to join.

Generated with Midjourney

Keep reading with a 7-day free trial

Subscribe to Stable Discussion to keep reading this post and get 7 days of free access to the full post archives.

Already a paid subscriber? Sign in
© 2025 Stable Discussion
Privacy ∙ Terms ∙ Collection notice
Start writingGet the app
Substack is the home for great culture