That is the question Fit2AI began with.
Today, a sports watch can record pace, heart rate, cadence, power, elevation, training load and recovery time. It can even estimate lactate threshold, race performance and training status. After an ordinary run, we may have hundreds or even thousands of data points, rather than just a handful of numbers.
We already have enough data.
What we lack is a way to understand it.
From recording training to understanding it
Most fitness platforms are very good at answering one kind of question:
What happened today?
How far did I run? What was my average pace, heart rate and training load?
But when we train, we often want to ask a different kind of question:
Why did it happen?
Why was my heart rate higher today than last week at the same pace?
The last two repetitions of this interval session were slower. Was that normal fatigue, or was the intensity too high?
After increasing my training volume over the past few weeks, has my threshold performance really declined, or is this a short-term fluctuation caused by weather, fatigue and the structure of my training?
With a race three weeks away, should I keep increasing intensity or start recovering?
A single metric usually cannot answer these questions.
They require us to place a session in a broader context.
That is where our thinking about Fit2AI began.
AI should do more than summarize a workout
Giving workout data to a large language model and asking it to write a few paragraphs of feedback is not especially difficult.
But we soon realized that if this is all we do, AI’s value is very limited.
A training session never happens in isolation.
The same 1km × 5 interval session means something very different for someone just returning to training than for someone who has completed three consecutive weeks of high-intensity work.
A five-second drop in lactate-threshold pace can also mean very different things in winter, midsummer, post-race recovery or a period of heavy training.
So what we really want to build is not an “AI workout summarizer.”
We want AI to see the context around training.
It should know what has happened recently, understand how sessions relate to one another and let users ask follow-up questions.
Gradually, the direction for Fit2AI became clear:
Turn training data into context for a conversation with AI.
You should be able to ask questions of your data
We do not want people to have to learn a new system of metrics to use Fit2AI.
You can simply ask:
“How did this session go?”
“Why did my heart rate keep rising in the second half?”
“Have I improved compared with the same workout last month?”
“Have I been training too much lately?”
“Given how I’m doing now, how should I adjust next week’s training?”
These are the questions runners already ask their coaches, teammates or themselves.
We want Fit2AI to bring together information scattered across workout records so AI can analyze it around these real questions.
That is why connecting data sources has become increasingly important to us.
FIT files, Apple Health, COROS and other sources we may connect in the future are more than just “import features” to us.
Together, they form the foundation for AI to understand training.
We are not trying to reinvent the fitness platform
From the start, Fit2AI was never meant to replace a sports watch or recreate Strava, COROS or Apple Fitness.
These products are already very good at recording and displaying workouts.
We are more interested in what happens after the data is recorded.
Once the data exists, what else can we do with it?
Can AI help people notice changes they have missed?
Can we understand a failed session differently by placing it alongside the past few weeks of training?
When someone is preparing for their next race, can the past few months of data genuinely inform their decisions?
Fit2AI is more like a layer for understanding, built on top of the existing fitness ecosystem.
Devices record.
Platforms store.
Fit2AI tries to make sense of it.
AI’s value is not in making decisions for you
Training is a complex system.
Weather, sleep, stress, physical condition, training history and even how you feel that day can change the meaning of a session.
So we do not believe AI should be an infallible “digital coach.”
Instead, we want it to be a tool for analysis.
It can spot trends, organize data, suggest possible explanations and help people reconsider their own judgments.
But the final decision should still belong to the person.
This is a principle we have held to throughout the design of Fit2AI:
AI should help people understand their training, not ask them to hand it over to AI.
Starting with a very specific problem
At LIGHTOUCH, we often start building a product with a very specific problem.
We do not first decide, “Let’s build an AI product,” and then look for somewhere to use AI.
Instead, we encounter a problem first:
We have plenty of training data every day, but when we really want to understand our training, we still have to search across different pages, compare the information and make sense of it ourselves.
AI happens to offer a new way to address that problem.
That is how Fit2AI came about.
It is still changing rapidly.
From analyzing individual sessions to connecting more data sources and giving AI a fuller picture of training, our understanding of this product keeps evolving too.
But the original question has not changed:
We already have so much workout data. Can we put it to real use?
Fit2AI is our attempt to answer that question.