I thought AI would build the app. Ha.

When I started building Thrive, I thought the hard part would be making an app.

The screens. The buttons. The flows.

And actually, that part moved surprisingly quickly.

Using AI tools, I went from an idea in my head to something that looked and felt like a real product much faster than I expected. I had user journeys. I had symptom and cycle tracking. I had a dashboard starting to come together.

For a little while, I thought: wow, maybe everyone is right. Maybe you really can just build an app with AI now.

How does it actually understand anything?

And that's where things got interesting.

A beautiful tracker is still just a tracker

I've been very clear with myself from the beginning that I don't want Thrive to become another app where you spend your day ticking boxes.

Pain: 7/10.
Sleep: 6 hours.
Exercise: yes.
Supplements: no.

There are already plenty of ways to record information.

What I want is something that understands the context around that information.

I want someone to be able to open Thrive and say:

“I had cereal and coffee this morning, forgot my magnesium, went for a walk, and now my left leg really hurts.”

And for Thrive to understand.

Not just:

User reported leg pain.

But:

What did they eat that day?
Did they take their usual supplements?
How did they move?
Where are they in their cycle?
How did they sleep?
Has this happened before?

Because I'm starting to realise that, for the product I'm trying to build:

The context is the data.

Which opened a much bigger can of worms

Once I understood that, my very satisfying phase of designing screens quickly turned into me learning an entirely new vocabulary.

Agents.
Data models.
Databases.
Prompt engineering.
Memory.
Orchestration.

Things I absolutely did not have on my maternity-leave bingo card.

It turns out AI can help you build an interface remarkably quickly.

But if you want to build something that actually understands a person over time, you need to think about what the system knows, how it stores that knowledge, how different pieces of information relate to each other, what it should remember, and what it should ask when information is missing.

That's a very different problem.

The Health Data Agent

So this is the part I'm working on now.

I'm starting to build what I'm calling a Health Data Agent.

The idea is that instead of making someone manually work through lots of separate trackers, Thrive should be able to understand normal language and organise it in the background.

If I say:

“I slept badly, had eggs and coffee for breakfast, walked for half an hour and forgot my supplements.”

I don't want to then have to separately open:

Sleep → Food → Drinks → Movement → Supplements.

The system should understand that I just told it five different things.

Eventually, I want it to do more than organise them.

I want it to recognise what keeps happening over time, connect information across different areas of someone's life, and ask useful follow-up questions when it genuinely needs more context — rather than filling in the blanks itself.

That's where the idea of Thrive as simply a tracker has started to feel much too small.

Why I care so much about context

One of the biggest influences on how I'm thinking about this came from somewhere I didn't expect: my acupuncturist.

I originally started seeing her for acupuncture.

But what I got was much more holistic than that.

She wanted to know about my sleep. My stress. What I was eating. When I was eating. How I was moving. What supplements I was taking. Where I was in my cycle. My digestion.

Sometimes she'd ask about things I would never have thought were relevant like the toothpaste I used.

Whether every individual thing mattered wasn't really the point.

What stayed with me was the way she thought.

She wasn't looking at endometriosis as a collection of isolated symptoms.

She was trying to understand the whole system around me.

And that's increasingly how I think Thrive needs to work too.

shout out to Alex O'Connor (Fertility Acupuncture Specialist & Fertility Mentor)

Endometriosis doesn't happen in neat little boxes

Living with a chronic condition certainly doesn't feel like filling out a database.

Sleep affects how I feel.

Certain foods affect my symptoms.

Pain changes how much I move.

How I'm feeling changes what I eat.

A treatment might seem useless after a week but look completely different when I compare three cycles.

And sometimes the thing that matters isn't one individual data point at all.

It's the pattern between them

That's what I want Thrive to eventually help uncover.

Not:

“Your pain was 7/10 on Tuesday.”

But something closer to:

“This has happened at a similar point in your last three cycles. Here's what else was going on around those days.”

That is a much harder thing to build.

But it's also much closer to the thing I actually wish existed.

So, no more screens for a little while

My instinct when something doesn't work is still to design another screen.

I'm learning that sometimes the answer isn't another screen.

Sometimes it's better intelligence underneath the ones that already exist.

So the next phase of building Thrive is going to look slightly different.

Less time obsessing over whether a card should move three pixels to the left.

More time figuring out how agents work, how health information should be structured, how memory should work, and how an AI system can ask questions without making assumptions.

I started this journey thinking:

I'm building an app.

I'm beginning to realise I'm trying to build an intelligent health companion.

Which is significantly more complicated.

But also far more exciting.