Three Weeks

The idea for Cellopack came to me at the end of 2021, and I built it through the first half of 2022.

At the time, most of my experience building things on the web was through WordPress. Lately I’d gotten deep into Webflow, fascinated by the no-code movement happening around it.

I could code, but I wasn’t fluent. It was a little like knowing enough Spanish to read it and understand what someone was saying, but not enough to comfortably have a conversation.

No-code let me work around that.

The first version of Cellopack was Webflow on the front end, Airtable as the database, Memberstack for accounts, Jetboost for functionality, and Zapier to move data between them. It wasn’t particularly sophisticated, but it worked.

In June 2022, I shared it with a local Facebook group and a few people in the hobby I knew.

Eight people signed up.

And then I kind of lost interest. I don’t think it was because only eight people signed up. I had built the thing. That was the part that had excited me.

I was also paying around $200 a month to keep everything stitched together. After six months or so, I had spent more than $1,000 keeping an experiment alive that I wasn’t doing much with.

So I shut it down.

When ChatGPT came along, it started filling in the gaps between what I understood and what I could actually produce myself. I could ask how a database should be structured or what an API was doing, get an answer, understand it, and keep moving.

By the summer of 2025, I decided to try Cellopack again. But I was still thinking mostly in no-code. I worked with ChatGPT to plan a more sophisticated version and eventually turned it into a roadmap that ran from August 2025 to April 2026.

Nine months to a full launch.

There was a lot going on around it. My wife had just gotten 15 chickens that were growing in my office, which meant I was building my first chicken coop. My three boys were in their first year of being homeschooled. I was joining my dad for his cancer appointments.

For a while, I had a lot of fun bouncing between projects. I’d go outside and figure out how to frame a wall, then come back inside and figure out how to structure a database.

The chicken coop turned out to be a much bigger project than I expected.

Life did too.

By December, my dad was really slowing down. Cellopack stopped mattering.

I didn’t touch it for a while.

A couple months after I lost my dad, I reconnected with an old friend. We talked about the things we’d each been trying to build, and he told me I needed to try Claude Code.

So I signed up.

Until then, ChatGPT could tell me how to build almost anything. It could write the code and walk me through setting things up.

But then I had to go do it.

Claude Code felt different.

I could tell it what I was trying to build and figure out what I actually needed. I’d create the accounts and connect them, and from there everything started talking to everything else. When something broke, we could work through it where it was actually happening.

It stopped feeling like:

Tell me how to build this.

And started feeling like:

Build this with me.

So I made a decision. No more no-code tools. I’m going to figure this out.

I pulled out the nine-month Cellopack plan I’d made the previous fall. And I started building.

About three weeks later, I had built the whole thing.

Not just the nine-month version either.

I built things I had intentionally left out of that plan because they felt too technical. A real backend. OAuth connections. Paid memberships. Identity verification.

Things I would have previously looked at and thought, I don’t know how to build that.

Three weeks earlier, I still didn’t.

AI made me faster, obviously. Turning a nine-month plan into three weeks is hard to describe as anything else.

But speed wasn’t really the thing that changed.

The nine-month roadmap had been written around what I believed I was capable of building. Other ideas had been pushed off until later because I didn’t know how to do them. Maybe I’d eventually learn enough. Maybe I’d find another tool. Maybe I’d pay someone.

AI didn’t just accelerate the roadmap.

It changed what I was willing to put on it.

By the end of May, Cellopack was ready to share. And I started procrastinating.

I’ve never been particularly afraid of building something that fails. If I spend three weeks building something and nobody wants it, I learned a lot in three weeks.

That’s a pretty cheap failure.

What I hadn’t expected was being afraid of the opposite.

What if 10,000 people sign up?

It sounds ridiculous now, but I genuinely thought about it. Could I support it? What breaks? What happens if this becomes bigger than I know how to handle?

For years, there had always been another constraint. Something I didn’t know how to build. Some technical problem I couldn’t solve. Some point where an idea became too complicated for me to keep going.

Suddenly those things weren’t stopping me.

I was.

So naturally, I procrastinated by building another entire product. This one was Maternal Revival, a platform for my wife Hannah’s birth practice. It’s still in progress.

Then I ran out of excuses. I launched Cellopack.

The original Cellopack had eight users and cost me around $200 a month to keep alive. Cellopack today is far more sophisticated than anything I imagined building back then.

The only thing I pay for now is Claude. A hundred dollars a month, spread across four projects I’m building and maintaining. Everything else is still running on free tiers.

I don’t think the lesson is that AI taught me how to code.

No-code gave me a way around the limits of what I could build.

AI gave me a way through them.

What changed is the distance between the thing in my head and the thing I can actually make.

I don’t have to know how to build the whole thing before I start anymore.

I can just start building.