I thought I was building a smarter health tracker. I think I'm actually building something else.

It's been a little while since I shared an update on Thrive with Endo, but a lot has happened behind the scenes.

A lot of the progress has actually come from things not working well enough and forcing me to rethink what I want this product to be.

One of the biggest examples is something that sounds incredibly simple:

Logging your day.

I've always wanted Thrive to make tracking effortless. If you're living with endometriosis, the last thing you need is another job. You shouldn't have to open five different trackers to record that you slept badly, had pelvic pain, went for a walk, ate breakfast and forgot your supplement. So I built Quick Log. The idea was simple: just speak or type naturally. "I slept 8 hours, I'm feeling good, had mac and cheese and took my multivitamin." Thrive would organise everything for you. Except… it didn't. It might pick up:

Sleep: 8 hours

and

Food: cheese

…and completely miss the mood, the "mac" part of mac and cheese, and the multivitamin. Another time, "toast with peanut butter" became simply "toast". Technically, these sound like extraction bugs. But they made me realise there was a much bigger product problem.

Humans don't experience their health in database fields

My database needs neat categories. Symptoms. Sleep. Food. Movement. Medication. Supplements. Treatments. Mood. People don't think like that. We say: "I barely slept last night, maybe five hours. Had my usual breakfast, forgot my magnesium and my stomach has been awful all morning." And another day we might simply say:

"Same breakfast as yesterday. Pain is much better today."

For a traditional tracker, that second sentence is almost useless.

For something genuinely intelligent, it contains a huge amount of context.

• What was yesterday's breakfast?
• What pain are we referring to?
• Better compared with when?
• Is this something worth recording?
• And does the person even want to track food?

That led to quite a big shift in how I'm thinking about Thrive.

I don't want to build a tracker that gets better at putting sentences into boxes.

I want to build something that understands the person first — and organises the data quietly in the background.

So we're building memory

The next version of Quick Log is becoming what I'm calling the Health Conversation Agent.

There will still be structured health data underneath. In fact, that's incredibly important.

But above that will sit a conversational layer that can understand context and, eventually, remember useful things about you.

For example, you might tell Thrive once:

"My usual breakfast is toast with peanut butter and coffee."

Later:

"Had my usual breakfast."

Thrive should know what you mean.

Or:

"When I say my pill, I mean [my contraceptive]."

Later:

"Forgot my pill today."

That shouldn't require another questionnaire.

And if you say:

"Actually, I did take it."

Thrive should understand that you're correcting what you just told it — rather than happily creating two contradictory health records.

Underneath, this means separating conversation, memory and health evidence.

The conversation helps Thrive understand you.

Memory holds useful reusable context: your terminology, routines and preferences.

And structured health events remain the reliable, dated information that can actually be analysed.

That distinction matters enormously when you're building something around health.

Because the other big piece is starting to work: understanding patterns across cycles

Over the past few weeks, I've also built the first version of Cycle Health Review.

Initially, I was building weekly summaries.

Then I realised: why am I forcing a seven-day reporting structure onto something inherently cyclical?

Endometriosis doesn't reset every Monday.

So the system now looks at a completed menstrual cycle as its primary unit of analysis.

It can look at things like symptom timing and severity, sleep, movement, mood, medication and supplements, treatments and interventions — and compare them with previous completed cycles.

But there is an important rule I've been quite obsessive about:

the AI isn't allowed to simply decide that something is a pattern.

The underlying system calculates the facts first.

And something can't be presented as "Repeated across cycles" unless the data actually supports it across multiple completed cycles.

That's probably not the sort of progress that makes a particularly exciting product demo.

But it's exactly the sort of progress I want to make before asking people to trust the insights they're seeing about their health.

The goal isn't to track more

This has also changed another assumption I had.

Most health trackers are designed around collecting more.

• More complete days.
• More categories.
• Longer streaks.
• Fewer gaps.

I'm increasingly convinced that's the wrong goal for Thrive.

Maybe someone wants to understand whether acupuncture is changing their symptoms over three cycles.

Maybe someone wants to track sleep, pain and movement.

Maybe someone takes several supplements and wants to understand whether anything changes after introducing one.

And maybe someone absolutely does not want food, weight or nutrition tracking anywhere near their health experience.

Their version of Thrive should reflect that.

If food tracking is switched off, I don't want the app saying:

"You haven't logged any meals this week!"

I don't want food appearing as an empty dashboard.

And I certainly don't want an AI insight pointing out that there isn't enough nutrition data.

For that person, food simply isn't part of the picture they've chosen to build.

Personalisation shouldn't just mean changing the recommendations. It should mean changing what the product chooses to pay attention to.

Where I'm trying to get to

Imagine opening Thrive and saying:

"Period started yesterday. Pain was really bad overnight and I only slept about four hours. Had acupuncture this afternoon though and I'm actually feeling much better now."

And that's it.

No opening Period.

Then Symptoms.

Then Sleep.

Then Treatments.

Then Mood.

Thrive understands what you said, asks one useful follow-up if something genuinely matters, and organises the rest.

Then, over time, it might be able to tell you:

"Your pain has consistently peaked during the first two days of your last three cycles."

Or:

"Since you started acupuncture, your logged pain severity has been lower in two of the following three cycles. There isn't enough information yet to know whether those things are related, but it may be worth continuing to watch."

And importantly, you can tap:

Why am I seeing this?

…and see the actual information behind the observation.

That's the product I'm much more excited about building.

Not an AI that gives you more health advice.

Not a tracker that demands more data.

But something that helps you build a picture of your own health with as little effort as possible and then helps you see things in that picture that are difficult to see day-to-day.

There is still a lot to build.

The Cycle Review needs more real-world testing. The conversational memory layer is the next major technical challenge. And I'm sure letting real people loose on it will expose assumptions I haven't even considered yet.

But I think the direction is becoming much clearer.

The more intelligent the product becomes, the less the person using it should have to think about tracking.

And that might be my biggest learning from building Thrive so far.