Building Minimum Viable Products using AI Tools - A Solopreneur's Adventure

Cursor, Bubble.io, Cosmic App Forge and Claude Artifacts

TLDR:

Buckle up, tech enthusiasts! I've been on a wild ride through the AI-powered app development jungle, and boy, do I have some tales to tell. If you’re not an engineer, don’t adjust your set. I have some options for you, too. Here's the scoop:

Claude Artifacts: Building apps on your phone? The future is knocking, and it brought snacks! Pros: Instant app creation, zero coding required. Cons: Still in early stages, limited customization.

Cosmic App Forge: The dark horse that surprised us all. From idea to MVP faster than you can say "AI revolution." Pros: Rapid prototyping, wide range of app types. Cons: Mysterious developer, potential security concerns.

Bubble.io: No-code platform with a dash of AI. Like LEGO for web apps, but you might need to phone a friend (or ChatGPT) for instructions. Pros: Visual development, extensive integrations. Cons: Learning curve, limited customization options.

Cursor: The coder's AI sidekick that sometimes forgets its own name. Powerful, but prone to "Dory syndrome." Pros: Advanced AI coding assistance, VS Code compatibility. Cons: Context retention issues and potential for refactoring rabbit holes.

Get ready, folks. The app-ocalypse is coming. Whether you're a coding wizard or a tech newbie, these tools are reshaping how we build the digital world.

Read On

I'll preface this Shep Report by saying I'm not an engineer. I just play one on TV. That is to say, I was an engineer very early in my career, and I even wrote code at Microsoft that shipped with Windows 25 years ago. But I quickly decided it wasn't my calling and hastily retreated to the much safer, less treacherous backwaters of Product and Program Management. Having said that, I code and tinker for fun, so not a complete neophyte.

I'm guessing the vast majority of readers probably aren't engineers either. . . but you may work with engineers, you may have had an idea bouncing around in your head for that next billion dollar company, or a side hustle that you know would be simple if you could get a couple of engineers, a designer, and some frothy AI angel investors to throw in some money.

As an entrepreneur and software guy who's worn most hats over the years and shipped and delivered hundreds of apps, experiences, and significant initiatives, I know what it takes to ship a product. It's damn hard work.

You need a founder or product manager who truly understands the customer and crystallizes that vision for the execution team to deliver the right product. You need a Program Manager or Scrum Master to beat the drum, pick up the myriad of random tasks, and ensure the engineers have all they need to code. You need a User Experience Designer who breathes life into the vision and crafts the perfect, frictionless magical experience, whether it's an accounting app or a 3D Digital Twin of a rocket. Then you simply, you know...wait for the engineers to do whatever they do. . . I kid, I kid. All of this pre-work is in service of helping the engineers manifest modern-day magic as they mutter their incantations and turn ideas into a working product.

It's expensive, dirty, dangerous work, and I love it. But. . . as an entrepreneur, the hardest part of the whole gambit is getting your first 1000 users, typically through a Minimum Viable Product or MVP, as it's commonly referred to. If you hit that milestone, you know there's a there, and raising money becomes infinitely easier. Once you've got that signal, you raise the confidence bar and can afford to hire that incredible talent described above without having to remortgage your home or raid your retirement fund.

And that's been the goal of some of my recent explorations.

As a technically savvy solo entrepreneur, can I build and test an MVP using AI?

I first wrote about this journey nearly a year ago (who knows how long ago that is in AI years) with my article, AI Solopreneurship - Experiments in using multiple coordinating AI agents to conceive, design, build, test and market a new business venture

Interestingly, my idea (generating images based on the paragraphs of a book) was implemented by Google Research last month with Infinite Wonderland. Even using the same example of Alice in Wonderland! I'm not saying they copied my idea. . . but maybe that sneaky AI did?

Adam’s (No-Such-Thing-As-An-Original-Idea App)

Google’s (Totally-Not-Related-App). . . probably

Pivot Time

So, with that, I shifted to a somewhat related idea: a productivity app that would assist busy individuals in visualizing their hectic day. I'm an enthusiast of Personal Knowledge Management Systems (PKMS) and productivity, and I use various apps like Notion, Obsidian, and Todoist to manage both my business and personal life. Much of it is interconnected with AI and routines that automate tasks as much as possible, providing me with a sense of organization (most days, except when I'm tweaking the system itself).

What I do know is that even when I look at a full calendar for the day, it's often quite challenging to visualize all the tasks, transitions, and the full scope of the things I need to accomplish to get through my work for the day.

What if an AI could analyze my to-do list and combine it with information about me, my life, my friends, family, and colleagues, and then visualize me doing those tasks in a vivid, linear, narrative style? It could help me make smarter decisions about how I organize my day and what tasks I can realistically tackle.

For example, realizing that fixing the sink involves multiple steps such as going to Home Depot (twice), getting tools from the garage, watching a tutorial, and explaining it all to a toddler, would provide a more accurate assessment of the time needed for the task.

Here’s some sample images I created manually to try and imagine the app using Flux trained on a handful of pictures of me.

Picking the kids up from school

Cooking Spaghetti Bolognese

Fixing a washing machine

So, I've decided to try and create a web app called "MindsEye" for this project. I've experimented with four popular AI-enhanced coding tools and I've gained a lot of insight into the current state of the art. While the pace of AI innovation is rapid—by the time you read this, some aspects may be outdated—but the issues I faced and some of the solutions will likely still be relevant.

The tools I tried are :

1) Cursor - The current darling of the Software Engineers AI toolbox

2) Bubble.io - A well-established "No / Low Code" tool with a big following for SAAS Web Applications.

3) Cosmic App Forge - So simple your grandmother could make an app and my current leading contender with some caveats. . .

4) Claude Artifacts - I just scratched the surface with this, but it is very promising as a simple app builder with zero engineering knowledge required.

I'll give a little synopsis of each tool below and try to avoid too much technical jargon, as there are plenty of resources online.

The starting point for most of these tools is to describe what you’re trying to achieve for the LLM in plain English. You can see the general prompt and response describing the application to be built here.

Cursor

What is it?

Cursor is a modified version of the popular Visual Studio Integrated Development Environment (IDE) that has been heavily customized to focus on AI-assisted programming. It maintains the familiar text editing experience of VS Code while adding advanced AI capabilities

Key Features

• AI Integration: Cursor is built around the concept of AI-assisted coding, offering features like next action prediction, natural language edits, and chatting with your codebase.

• LLM Support: It works with various large language models, including Claude 3.5 Sonnet and GPT-4.

• VS Code Compatibility: Cursor supports the full VS Code plugin ecosystem, allowing users to install and use their favorite extensions

My Experience

I was initially impressed with Cursor. I primarily used Anthropic’s Claude Sonnet 3.5 within Cursor, as this was the top coding LLM at the time of writing. It's a powerful tool for more capable engineers. I especially liked that, instead of sharing code snippets in the chat with the LLM, I could have it analyze and review the entire code base it had generated.

BUT. . . this was also a major issue. Because I was doing so much back and forth in describing what I wanted the app to do, it very quickly could no longer hold the breadth of the code, my descriptions of the app, and previous decisions we had made together in its context window. Essentially, it was like coding with Dory.

The AI and I would go down a rabbit hole of trying to tweak or fix some minor issue, and then it would try and refactor and rebuild the entire code base, inevitably breaking something. This quickly became an exercise in frustration. I also hit a limit of usage for the day with Claude, so I decided to switch to ChatGPT 4o. This started producing better code and overall results for me, but I found that I had to try and summarize the whole multiple-hour-long conversation with Claude to ChatGPT and ended up in the same issue of refactoring the entire code base. Who knew you couldn't just hot-swap out an AI engineer for another without issue?

I don't wholly blame Cursor / Claude / ChatGPT for this. It was a mismatch of expectations, and some mitigating strategies can be employed to improve the experience

However, for my limited use case, I decided that Cursor may not be the solution I needed for a quick and simple MVP.

Bubble.io

What is it?

Bubble.io is a powerful low/no-code platform that enables users to create web applications without writing traditional code. It offers a visual programming interface that combines design, logic building, and database management in one comprehensive tool.

Key Features

• Visual Development Interface: Bubble.io provides a drag-and-drop editor for designing user interfaces and building application logic visually.

• Database Management: The platform includes built-in database functionality, allowing users to create and manage data structures without setting up separate database systems.

• Workflow Engine: Bubble.io features a powerful workflow editor that lets users define complex logic and actions using a visual, sentence-based approach.

• Responsive Design: The platform supports creating fully responsive web applications that adapt to different screen sizes.

• API Integration: Bubble.io allows seamless integration with external APIs and services, expanding the capabilities of applications built on the platform.

• Scalable Hosting: Applications built with Bubble.io can be deployed and scaled easily, with hosting handled by the platform itself.

My Experience

With some searching on Reddit, I found that Bubble.io was very well-established and, even before the AI Era, had been a popular choice for Web Apps and SAAS experiences such as:

- Custom marketplaces (similar to Uber or Airbnb)

- Social networks

- Booking applications

- Internal management tools

- Job boards and directories

I quickly pulled together a basic Todoist clone to allow me to create tasks. I felt more in control of the user interface and environment when developing the app through its drag-and-drop editor.

In this case, there's only some rudimentary AI integration, and for the most part, it was me cobbling things together. However, I did use ChatGPT 4o pretty extensively to help troubleshoot and debug as I hit the limits of my knowledge with Bubble.io.

I discovered that the most effective way to help ChatGPT understand my goals was to take screenshots of the user interface and database schema and then ask for detailed, step-by-step instructions on what to do next. While this method was somewhat effective, I faced some challenges as its knowledge base was outdated due to changes in Bubble.io since the language model was trained. Additionally, the responses tended to be too high-level when what I needed was simple, step-by-step instructions on what actions to take within the application.

In summary, with some more effort, Bubble.io could be a good MVP testing tool for me, but it’s still somewhat laborious and doesn't have the level of AI integration I'd like.

Cosmic App Forge

Finally, I was browsing some subreddits and found a post for something called Cosmic App Forge from a Finnish developer. I clicked the link, was a little put off by the URL being https://aidark.net, but proceeded anyway (do so at your own risk...I believe it's legit) and started describing my app.

What is it?

Cosmic App Forge is an innovative AI-powered application development platform that aims to revolutionize the way apps are created. It leverages advanced AI technology, specifically Google's Gemini AI or ChatGPT4o, to enable users to build complex applications with minimal coding knowledge.

Key Features

• AI-Assisted Development: Users can describe their app ideas in natural language, and the AI generates functional code based on those descriptions.

• Wide Range of App Types: Cosmic App Forge can create various applications, from simple portfolio websites to complex full-stack social media platforms.

• Integration with Popular Technologies: The platform can work with technologies like Google Maps, Firebase, and various mobile platforms.

• Rapid Prototyping: Users can quickly create and iterate on app ideas, significantly reducing development time.

• Cross-Platform Development: The tool supports creating apps for different platforms, including web and mobile.

• Advanced Capabilities: It can handle complex tasks like creating augmented reality apps or multi-user collaborative tools.

My Experience

I have to say I was impressed. With an elementary description, I had a functional MVP up and running in about an hour. The images are generated by OpenAI's Dall-E, and all I had to do was give it an API key. While I would likely use Flux.ai to create images trained on my likeness in the future, this was a great proof of concept that took me significantly less time to build than the other tools.

I will say I'm a little hesitant to thoroughly recommend it for serious development, mainly because the developer's identity is somewhat obfuscated, and I don't feel 100% secure providing things like personal info and API keys to an unknown 3rd party.

However, the risks/rewards in my particular case were worth some limited risk-taking to get results that I was happy with. Your results may vary, and again, you may do so at your own risk. It's the Wild West of AI right now and so if you're going to explore the cutting edge, expect to get 'cut' now and again.

Conclusion

I'm happy to have ramped up on some of the more popular and new tools and techniques for app development in AI. While I haven't finalized and deployed something to real customers quite yet, I'm optimistic that with a bit more work, I can get to an MVP that I can test with real people.

One of the folks who’s probably at the forefront of solopreneurship is @levelsio, who has created many MVPs and is very generous with his knowledge and experiences online. Well worth a follow if you’re looking for further inspiration.

To me, despite these tools being clunky, rudimentary, flawed, and, let’s face it, complex, it's a sign of things to come. I don't believe engineers are going to go the way of the Dodo, but I do think how they work and what they work on is going to change dramatically. I also believe in the same way the web empowered billions of people to become content creators, the same will happen with applications. Everyone will have custom apps, services, and agents that will be highly tailored to their individual needs and preferences, as I discussed in "The Beginning of the end of applications as we know them."

I'm excited to be a part of shaping this new future. Putting in the hours to roll up your sleeves and figuring out what works and what doesn't work is part of the voyage of discovery and helps to separate the hype from reality.

Thanks for taking the time to read. If you’re looking for help with your AI or Emerging Tech Strategy or have an exciting leadership role you’re looking to fill, I’d love to hear from you. You can contact me directly here.