The challenge
I want to see if I can build a successful micro-SaaS from idea to revenue in one month using Claude Code as a non-technical entrepreneur. But I don’t just want to do it, I want to document the journey and help others do the same. I have no idea if this experiment will work but fear of failing publicly is never something that has stopped me in the past so I’m making a gamble here and attempting true real-time #buildinpublic on this quest. I love this vision of hacking through wild undergrowth to figure something new out and then once you do, laying pavers to make the path easier for anyone else who wants to follow. This is in the spirit of this whole Stone Soup Vision and I think this will be a good exercise.
Some background
I’ve been using Cursor AI to build various little tools and apps since around last Christmas. I also run the Nocode Lisboa meetup in Lisbon and have been demo’ing how to use tools like Cursor, Loveable and Bolt for creating MVPs and internal tools (you can find recordings of all my talks on each of the event pages there). I have been hearing how much better Claude Code (CC) is than Cursor so I finally knuckled down this past weekend and started tinkering with it.
I am not a command-line guy. I grew up with a mouse, I think visually, I shudder at terminal sessions and Linux commands so the furthest into the technical waters I’ve ventured thus far is the IDE experience with Cursor. The fact CC has only a terminal-based UI is the thing that has kept me from trying it but recently VS Code released an extension that allows you to run CC within its editor. This makes it possible to run CC within the terminal window of VS Code (of which Cursor is just a fork) so it got me to finally try it out.
Wow. Just… Wow. It produces hands-down significantly better output with a fraction of the headaches. Cursor is great- don’t get me wrong, it’s amazing – it’s just that CC is next-level. I don’t understand exactly why it produces fewer mistakes and seemingly understands the goal-directed prompting better because I was previously using the Claude Sonnet 4 model as the model in Cursor. But it is objectively better.
Getting Real
The best way to learn an instrument is by developing the desire to play a song, and then tinkering until you can make the song. The best way to learn a tool is to cultivate the desire for a product and then pursue building said aspirational product with the tool. I’ve had this idea for a little utility that could take any form and convert it into an interactive chat experience (much like how I used TypeBot to make the comments of this blog interactive). But I want to make this to work for every major form provider out there:
- Google Forms
- TypeForm
- Survey Monkey
- SurveyJS
- Gravity Forms
- WooForms
- etc
And I want it not only to convert the form into an interactive chat session like TypeBot (because that’s relatively easy to do) but I want it to become knowledgeable of the form owner’s goals and have a more in-depth conversation and extract valuable insights as well as the structured, validated form data from the respondent. The agentic AI frameworks are advancing quickly and this seems like a candidate for a useful service that could be built while yielding the opportunity to play with these bits of tech which are all evolving rapidly:
- Claude Code
- MCP Servers
- Agent frameworks like Mastra.ai, Crew.ai, LangGraph and comparing those with what I already know in n8n
As a bonus I see it as an opp to play with this other stuff I want to learn:
- basics of Git, Github and doing proper source control in teams
- CI/CD pipelines and automations for proper dev/test/staging workflow and bulletproof deployments
- NextJS, Supabase, DrizzleORM, Clerk for auth
- Linear for project management and task tracking
I have no idea how this is all going to turn out but I’m three days into this experiment and thus far I’ve been able to successfully create a proof of concept here called AgentForm.app
Current Status
Here’s a quick Loom video of what I have so far after three days:
No promises but I’m going to push the limits of how autonomous I can make Claude operate and see if I can work with it to deliver a revenue-generating micro-SaaS product by end of month. Worst case it fails as a product but it’ll be if nothing more a useful experiment in learning all the above tech. I will share my lessons and takeaways here as I go so if you want those subscribe below.
NOTE: I am not a developer. I can’t write code at this point to save my life. I’m coming from the Bubble.io world of nocode (our entire system for Problemattic is built in Bubble and I’m giving serious thought to scrapping it and rebuilding it all with these tools). I was once a programmer way back in the day (’97-2004) so I understand the basic ideas but I am absolutely worthless at writing code at this point. My thesis though is that the quality of the models has evolved sufficiently to finally support the dream of being able to operate entirely as a product manager and cooperate with the AI as if it were CTO/Engineer. At any rate I invite you to follow along with this experiment by adding your email below. If you have something in particular you want me to document please leave a comment below. cheers
Thanks for reading. Could you answer a few questions below? This is incredibly helpful for understanding my readers and helps me create better content. thanks so much. -Sean
