Note ·
How I build software with AI agents
I build software by deciding what the product must be, writing that down, and then working alongside AI agents as they build it. I taught myself to code this way over the last year, and FlockConnect would not exist without it. The direction stays with me. I guide the AI. I do not let it guide me.
How did I start?
I started with a prototype built on Lovable, a tool that turns plain-language prompts into a working app. I was not a trained engineer. A mentor told me I no longer needed an engineering team to make a prototype, so I should build something I was passionate about. I did, and I put it in front of pastors.
Dozens of pastors looked at that prototype and told me what they needed. Their feedback shaped almost everything. Some of the most useful feedback was the hardest to hear: it was too complicated. I had built a lot of moving parts before I knew which ones mattered.
Why did I start over?
I started over because the first version could not carry what a church trusts it with. Lovable was good for a prototype. When I moved the code into Claude Code and tried to refine it, the architecture underneath kept fighting me. It had been pieced together, held up with duct tape, and church member data deserves better than that.
So in May 2026 I paused. I wrote the specifications and the architecture properly, then began again from scratch. I also told the churches already using the first version that it was not good enough, and that I was taking it offline to rebuild. That was a hard conversation and a scary choice. It was still the right one.
It felt like I had wasted a year. I had not. Everything I learned from those pastors and that prototype went into the second build, and the second build has gone far more smoothly.
What does working with an agent look like?
Working with an agent means the most important work happens before it writes anything. Give an agent a vague prompt and it will build something. It might even look glorious. It probably will not be what a church needs. So I write the specs first, then walk alongside the agent as it builds and keep checking its work against them.
The agent is fast, and it does not get tired. It also does not know a single pastor. It has never sat in a small group or wondered why a friend stopped showing up. That knowledge has to come from me, and from the pastors I keep talking to, or the software drifts toward whatever is easiest to build.
What would I tell someone who wants to build for the church?
Ask two questions before you build anything for the church: does the church actually need this, and does it already exist? Just because you can build it does not mean you should. Then build in the open. Show pastors what you have early, and let their answers change the plan.
I expected to be further along by now. I also know that two years ago, without these tools, I would not be anywhere close to where I am. Both of those things are true at once.