BubSync started at 3am in a hallway in Brisbane. Sarah was on maternity leave and I was back at work, so we split the nights.

The trouble with splitting a night is that I’d be up at 3am with no idea when our daughter had last fed. The one person who could tell me was asleep down the hall. So I guessed.

Then we left her with her grandparents for the first time. We wrote out a page by hand. Then I stood on the doorstep and gave them the spoken version anyway. Last feed, last sleep, last nappy. They had raised children of their own. They didn’t need my instructions. They just needed the last six hours.

That’s what we built. Not the logging, the handover. Every app in this category records what happened to your baby. Nobody sells the moment the baby changes hands and everything in one person’s head has to get into another person’s. That’s where the mental load actually lives.

So everyone who looks after your baby gets their own login instead of borrowing yours. One subscription per family, never per person, because a per-person fee is a charge for asking for help.

Nobody should need your password to look after your baby.

At the time of writing, BubSync is still in open testing ahead of our 28 August launch, and every tester had come from one of us asking someone directly.

Search hasn’t gone how I expected. I built the site for it properly. That isn’t the game. The bottleneck is backlinks, and you can’t engineer your way out of that. Google had indexed one page of ours out of thirty five, not because the pages are bad, but because nothing on the internet points at them. AI will audit your site and draft your copy. It can’t make another publication vouch for you.

I had traffic wrong too. Traffic is easy. The right traffic is hard, and the gap between the two is most of the work. So I build things I can give away, nine free tools on the site now.

Where I want it to come from is the product. Every parent who invites a grandparent or a carer sends us someone with a real reason to install. Invite to install is the number I watch, because it’s the only channel that gets cheaper as we grow.

I handed over the code. All of it. Around 266,000 lines of TypeScript in 775 merged changes since I started tinkering in November. I’m bootstrapped, so this was never a productivity gain. It was the difference between BubSync existing and not existing.

The hard part wasn’t the code. It was building the thing that checks the code. About 117,000 of those lines are tests. Once you can write code faster than you can read it, your only defence is a machine that checks the work harder than you would. My job stopped being "write software" and became "design the system that decides whether software is good enough to ship".

What I kept is judgment. AI can confirm a screen is accurate. It can’t tell me something is slightly off, and slightly off is most of the work in an app you open at 3am with one hand. The technical design is the same. AI will build whatever you point it at, including the wrong thing, and a bad decision there quietly makes everything after it harder.

Both are taste, and taste is the part nobody hands over. But the steering compounds. Keep making the call and explaining why, and eventually it starts picking up what you’re putting down.

Yes, and you can check it in ten seconds. Ask Google or ChatGPT what BubSync is. It’ll tell you. That isn’t luck. I built the site for answer engines early, which mostly means being structured and quotable.

Here’s what surprised me, and it cuts against what I just said. Backlinks are still a wall for ordinary ranking. The AI channel isn’t like that at all. One page of ours, a piece on 2026 baby trends, has been cited 200 times. That page was, for a while, the only page of our entire site Google had indexed.

The two channels have completely different barriers. A blue link ranking is gated on authority you build over years. Getting quoted in an answer is gated on whether you’ve written something a machine can lift, and a brand new site can do that on day one. I won’t oversell it. A citation is not a visit, and sometimes that works against you. People rarely click the citations when the answer is given first.

We haven’t launched, so I can’t tell you what people already know. What I’d say is the first impression increasingly isn’t your website at all. It’s a sentence about you inside somebody else’s answer, written by a machine, and you don’t get to edit it.

Two things to do. Spend as much time on the business as you spend on the product. Building is the comfortable part, and you can hide in it for months feeling productive. That’s the one I’d go back and tell myself.

And build the checks before you build the speed. Generating code is cheap now. Reviewing it isn’t. Skip the tests and the gates and what you have in twelve months is a very large codebase you’re quietly frightened of. That fear is the real failure mode of AI assisted development, not bugs.

Two things to ignore. The first is AI as your point of difference. In our category the market leader ships an AI assistant and a dozen small entrants lead with it. Nobody chooses a baby app because it has AI in it. The payoff isn’t building AI products, it’s building products with AI.

The second is your own self doubt, and it’s the one that will actually stop you. It creeps in when you’ve done an enormous amount of work with nothing external to show for it. Most people who don’t make it aren’t beaten by the market. They talk themselves out of it before the market gets a say.

The thing nobody warns you about is the number of hoops. Store submissions, business structure, accounting, terms, privacy. AI helps, but a human still has to sign and click and hold the account in their name.