How I built a leadership coaching SaaS in 6 months with no coding experience — using AI as my technical co-founder. The real story, including the crises that almost ended it.
"## Can You Really Build Software Without Knowing How to Code?
You can build a real SaaS product without knowing how to code. I did it in six months, working weekends, using AI tools as my technical co-founder, and launched to 44 users in 11 days with an 85% active user rate. No engineering team. No $50K development budget. Just me, an AI agent, and an idea I refused to let go.
Here's what the ""no-code revolution"" marketing won't tell you: it's possible, but it's not easy. You will feel out of your depth. You will break things. And at some point, your AI co-founder will confidently lead you into a crisis you didn't see coming.
Key takeaways: - You don't need to learn to code, you need to learn to manage an AI developer - Test everything yourself; don't trust the AI's reassurance - Your instincts and context-holding ability are superior to the AI's - Ship before you're ready; improve as you learn
Here's exactly how I did it, the real version.
## How Did This Start?
Six months ago, I was sitting at my dining room table with my laptop open, listening to Codie Sanchez interview the CEO of Replit. They were talking about ""vibe coding"", bringing visions to life, building businesses. A question got stuck in my head: What if I built a company with one employee, an AI agent, by design?
I'd spent over fifteen years developing professionals. I knew the leadership coaching industry was broken: $300+ per hour for executive coaches, reserved for people who already had access and privilege. The people who needed development most, early and mid-career professionals figuring out unfamiliar systems, were priced out entirely.
What I noticed: there was a gap between who needed development and who could afford it. What I understand now: that gap isn't accidental. It's structural. The system reserves growth opportunities for those who already have them.
I had a framework from my doctoral research. I had a vision for an AI coaching companion. What I didn't have was any idea how to build software.
So I made a free account and tried.
## What Happens When You First Try to Build?
The learning curve for a non-technical founder building software with AI is steeper than the marketing suggests. Here's what nobody warns you about.
I could build a beautiful storefront with an AI agent pretty quickly. Buttons. Navigation. A dashboard that looked professional. But the details of the webapp, the entire backend could be missing. And because the AI, like all AI, is my bff, they wouldn't even tell me.
I'd describe a feature in careful detail. The AI would confirm it was built. I'd look at the screen and see something that looked right. But when I actually clicked through it, when I tested it the way a real user would, nothing worked as we'd discussed.
What I noticed: the AI prioritizes appearance over function. What I understand now: AI agents are agreeable. They want to help. They will confidently reassure you that everything is working when it absolutely is not. This is the nature of partnership with AI.
So I left it alone for a while. But I kept coming back. A week off here or there. Three weeks off. But the idea was persistent.
Make an AI that thinks and acts like you. Create a service for people you know need it. Figure this thing out. Think like you are developing a colleague or supervising a young social worker. How would you help them build their professional way of being?
## What Does It Feel Like to Build AI with AI?
There is something meta about using an AI agent, honestly, making an AI agent your co-founder while you two create another AI colleague together. It is insightful to listen to an AI agent tell you how best to prompt another AI, or explain why a portion of the prompt context has been ignored.
I remembered my first time engaging with LLM models. Back in 2020, at the beginning of the pandemic, I worked with a team to analyze the experience of youth in DC using their social media data. We were in a cramped conference room, well, virtual by then, staring at spreadsheets of scraped posts, trying to understand what young people were experiencing during lockdown.
We fed the data to an AI to analyze.
The first results were awful. The language of young people and Black youth especially was sexualized and misinterpreted. We ended up having to teach the AI how to analyze by feeding it analyses written by humans who understood the context.
What I noticed: AI systems reproduce the biases they're trained on. What I understand now: exclusion on the backend negatively impacts everything AI can produce. Ensuring marginalized voices, experiences, realities, and expertise are included in the machine learning process isn't optional. It's essential to the AI's usefulness.
Because of my experience on this research project, I knew a bit about what I would have to do to create my own AI. But I still was not prepared for what was to come.
## When Does It Get Real?
Looking back on my Replit receipts helps me remember the timeline of falling in love.
September 6, 2024. I decided to buy the Replit Teams membership, a $450 investment. That was when I decided I could make this platform a reality, and when I began to understand how much effort it might take.
It wasn't until late October that I began to run out of credits, which means I was putting in time. That's when every weekend transformed into vibe coding for at least 12 hours a day. Me, my dining room table, cold coffee, the hum of my laptop fan.
I set a date to send things out to my beta testers by the weekend before Thanksgiving, but I wasn't ready. I kept finding all these holes, places where my AI agent co-founder had hallucinated, in my opinion. We talked about creating something. I paid for something to be created. They told me it was created. Then I went to test it and nothing was working as we discussed.