Steal my Idea: Stack Overflow for AI Prompts

What's up, everyone? This is Alex, the host of Founders Journal. As you may know, I built Morning Brew from my senior year of college until we sold the business in 2020. And now I'm building a variety of businesses in different spaces and different stages. I'm basically averaging 1 to 2 new businesses per year. And because of that, I have had the opportunity to test a ton of different products and services, ultimately landing on a core software stack that serves as my business operating system whenever I start a new company. I have my go-to banking service. I have my go-to formation service. I have my go-to domain hosting service. But probably the most important of them all is my HR service, Gusto. If you're a longtime Founders Journal listener, you know how much I think about people in the context of business. A business is nothing without its people, which means a platform that helps me manage and take care of my people is invaluable. And that means I've tried every HR platform under the sun, always with the same exact goals in mind. First, I want a product that's intuitive and delightful to use. As a non-HR person, I find the intricacies of payroll, benefits, taxes, compliance, and expenses to be painful. So I just want something that makes my life easier. Second, I want it to be affordable.

I'm a bootstrapper at heart.

I started Morning Brew with $0 and I started StoryArc with $0 as well, which means I don't have infinite dollars to drop on SaaS products. And third, I want a platform that acts as my one-stop shop for all things people. I want it to handle payroll, benefits, hiring, onboarding, and taxes, all while integrating with all of the other tools in my business stack, whether it be QuickBooks, Slack, or Google Workspace. Well, Gusto hit the trifecta, which is why I am a customer and use it for every single one of my businesses. Yes, this is a paid sponsorship by the amazing team at Gusto, but I would sing their praises even if I wasn't being paid to do so. So if you're looking for a platform that takes the pain out of payroll, benefits, and HR, I highly recommend you check out Gusto. And if you sign up at gusto.com/alex, you will get your first 3 months free. Now onto the episode. What's up everyone?

This is Alex Lieberman and this is Founders Journal. For those who are longtime Founders Journal listeners, welcome back. For those who are new to the show, here's the deal. Founders Journal is my audio diary made public for the world. My goal is to increase your odds of success as an entrepreneur by sharing the stories, lessons, and tactics that I experience as I build my businesses. You can expect the podcast Monday and Wednesday with episodes that are under 20 minutes in length. And if you want to get notified when a new episode comes out, make sure to subscribe wherever you listen. Today on the pod, I'm introducing a new format called Steal My Idea. How Steal My Idea works is I will pitch a business idea to you that I think could turn into an 8-figure business. Why would I do that? You might be wondering. Well, first, I have more business ideas than I have time or energy. So ideas that I'd love to see exist in the world that I think can change an entrepreneur's life I want to share. Second, even if you don't steal the idea, I think the act of thinking through a business idea and putting it through the idea maze is a valuable skill in itself. Lastly, before we hop into the idea and the episode, if you haven't yet introduced yourself, please do so. I love meeting Founders Journal listeners, so shoot me an email at alex@storyarb.com, say hi, and I'll get the conversation going. Now let's get into it. Today's Steal My Idea is all about Codex, which is Stack Overflow for AI prompts. Codex is a platform where users and companies can find, share, and create the prompts that make AI work better for them. Let me start with some context on how I arrived at this idea. I have stumbled upon 2 truths within the world of AI prompts over the last several weeks. First, the quality of your prompting determines the experience that you have with AI or an LLM. This became crystal clear as I started working on Distro, which is my AI content interviewer. Before we refined our 2 core prompts, the first that informs the way in which Distro interviews you and the second that controls how your interview transcript is turned into social content, the product was mediocre. The conversation with Distro felt flat and the content drafts felt like AI slop. After tuning the prompts, the product truly went from a B-minus experience to an A-minus experience. The interview felt more human. Distro felt less like a parrot and more like an intellectual sparring partner, and the content drafts felt 75% ready to post. Second truth: the time and energy required to go from bad prompts to great prompt can be tens to hundreds of hours. Just using the example of Distro, we probably spent 15 hours tuning our 2 system prompts to improve the quality of the product. Imagine if rather than spending those 15 hours fine-tuning, I was able to copy someone else's prompt that was 90% of the way there and spend just an hour or 2 fine-tuning it for Distro's needs. That was the genesis of what I'm calling Codex, like an ancient book of secrets, which is what a codex is, hence the name. This business will give users the key they need, in this case, the best possible prompt to harness the power of the LLM they are using. Now, I want to give you the quick elevator pitch for the business as if I was pitching an investor, and then I want to go through some of the questions or objections that they may have and how I would handle them. Let's do it. Today, using AI well depends almost entirely on how good you are at writing prompts, but crafting the right prompt is painful, slow and full of trial and error. Gary Tan, the CEO of Y Combinator, recently shared the 2 most important things in the world from here on out is no longer what school you went to or what your resume looks like. It is the amount of agency that you bring to the table and the taste you have that lets you get the most out of AI. Agency means prompting and taste means evals or evaluating how good the AI output is. Said another way, prompting is how society will unlock the full potential and power of AI. And our company Codex is on a mission to make sure that happens. We're building a platform where AI users can find, share, and create high-performing prompts powered by community voting, simple categorization, and expert curation. Our core insight is that prompting has an effort-to-effectiveness issue. A great prompt can be the difference between AI unlocking 10x results in your work or falling flat. But getting the right prompt can take hundreds of hours that you likely don't have. Codex is the solution to the problem. Think of it like Stack Overflow for AI prompts. I am the right founder for this because I'm an AI maximalist living the user pain daily. I have deep channel relationships with potential distribution and acquisition partners, and I am a community-oriented marketer at heart. Similar to Stack Overflow, which does $100 million in ARR and was bought for almost $2 billion, we will monetize primarily through enterprise SaaS, helping companies manage, share, and optimize their prompts internally. And then we'll have additional revenue upside coming from marketplace and advertising monetization. The initial market for this product is 91 million daily AI adult users in the US alone, and it is growing rapidly. Long term, we believe this can become a core acquisition target for foundation model companies, whether it's Anthropic, Perplexity, OpenAI, any of those that are looking to own the interface layer of AI. That's the pitch, which, by the way, I had ChatGPT help create most of it. Now, the idea may sound great to you, but as important as establishing the pitch and the business idea is steelmanning all of the unanswered questions and potential challenges that we will face with the business before deciding to pull the trigger and actually build the thing. So I went back to ChatGPT and I asked it, based on the pitch, what questions would a world-class investor like Marc Andreessen or Peter Thiel ask after I have given the pitch and what will be their biggest concerns with the idea? Here are the 6 questions they may ask me and my initial responses. First, how does this become defensible over time? If prompts are relatively easy to copy, how do you build lasting moats? My thought is community-based businesses are highly defensible. There are network effects where for every person added to the community, the product becomes that much more valuable. Actually, the hardest part with this type of business is the cold start problem of establishing the network, which in the case of this company, I just think you need to have tons of prompts ready to go and a core team who will upvote, downvote, and contribute prompts in the early days before you have critical mass. Question number 2: What is the biggest behavior change that you're betting on with this business? Are people really willing to curate, vote, and comment on prompts like they do for posts or code? My thought is the biggest behavior change is that more people will adopt prompting as they begin working with LLMs and AI tools, and that this change will stay. Which leads into the 3rd investor question. Question 3 is: Why won't a foundational model just eliminate the need for good prompting? What if in 12 to 24 months, AI gets so good that it can autocomplete bad prompts perfectly? It can self-clarify what the user meant. It can optimize its own internal prompting without external human help. Now, my thought here is, this is possible, and I don't claim to know where AI will be 12 to 24 months from now, but I have a thought on why this may not be where things go and what will happen if I'm wrong. My prediction is prompting will stay around because when a user prompts, they are working with limited information. The only way for them to refine and improve a prompt is to see the output of what that prompt creates for them or what it creates for their customer, which gives them more information to work with and helps them understand where their initial assumptions of what their prompt should have been were actually wrong. I don't think AI will be able to totally predict the quality of output or someone's reaction to a prompt. But let's say I'm wrong and prompting is rendered obsolete because LLMs can identify when a prompt is bad and how to improve it proactively. My hope then is that enough time will pass before this actually happens, such that you've built up a trusting, at-scale community of AI-first users who will need some other problem solved for them if this point comes in the future, and even if that problem is no longer finding the right prompt. Question number four: How fast can you build a dominant community before incumbents react? If OpenAI, Anthropic, or Stack Overflow notices you at six months traction, are you already too far ahead? So there's a few reasons I'm actually not too concerned about an incumbent trying to enter this business. For OpenAI and Anthropic, this isn't their core business, and building one great business is hard enough. Second, what may look like a big opportunity to me in the future looks like a tiny opportunity to. Incumbents today. And third, even if they do choose to build a similar business, it is hard to become known for a new thing like the go-to database for AI prompts when you are already known for something entirely different. Not impossible, but very hard. Question number 5, who is the first customer that you will close for your SaaS product and why? My answer is, given my relationships with OpenAI, Anthropic, and Perplexity, I'd work to strike up deals with them first. And the last question: Why does this need to be a venture-scale company? This is actually a great question, and honestly, I don't know that it has to be a venture-scale company. This could be a great candidate for seed strapping, where we raise a small $500,000 to million-dollar round after having some semblance of product-market fit in order to build out the core team that will allow us to grow. The best path for this business is likely to raise one round. Stay profitable, and then make the decision down the road of exiting the business to one of the foundational model companies or just cash flowing the business as a lifestyle business for a long time beyond that. So that's it for Steal My Idea. If you like the format, let me know at alex@storyarb.com. And if you have an idea that you want me to take through the idea maze like I just did for Codex, let me know and I may pick it for a future episode. That's it for today's Founders Journal. Thanks again for listening, and I'll catch you next episode.