From Stone Tablets to Silicon: The Computational Pathways of Knowledge in Coaching

How AI and LLMs Are Redefining the Coach’s Role in the Age of Data Overload

“Three-minute all-out on the bike,” Andrea said. No instructions, no explanations. Just ride.

So I did. And when I returned home, there it was - a notification from the AI Coach on my phone: “New 3-min power personal best detected.” No download, no manual analysis, no fiddling with files. It just knew.

This moment—simple, fast, seamless—is the beginning of what I believe will define the next chapter of coaching.

​

Why the Cognitive Burden on Coaches Is Unsustainable

Coaching has never been more complex. We're flooded with data: power files, GPS tracks, HRV trends, wellness scores, sleep reports, session RPEs. All potentially meaningful. But human capacity is limited. Decision fatigue is real.

When I was a program lead at High Performance Sport NZ, our sport scientists would spend entire days slicing files and squinting at charts just to uncover a training insight. We needed a better way. That search eventually led to Dr. Andrea Zignoli and the concept behind our recent paper: "The Computational Pathways of Knowledge in AI Coaching."​

This work shows how AI agents and large language models (LLMs) don’t just mimic coaches—they compute like them. They form semantic trajectories through a space of knowledge, navigating from first principles to context-specific prescriptions. This isn’t just a chatbot trick. It's reasoning at speed and scale.

And it begs the real question: what should the coach of the future do?

​

What LLMs Really Do for Us

Many coaches still see LLMs as glorified search engines. That’s a miss.

Imagine never having to dig through our 625-page ​HIIT Science​ book again to find the right interval prescription for a hot-weather taper week. Instead, you simply ask, "What session is best for this athlete, right now?" And the answer appears—grounded in foundational science and tailored to the individual's recent data.

LLMs can blend structured data (like training loads) with unstructured data (like an athlete's journal reflections) and deliver context-rich insights. This isn’t just about numbers anymore. It’s about making sense of the athlete’s full lived experience.

Delegating vs. Remaining Intensely Human

Let me be clear: coaches are not going extinct. But we are being redefined.

Here’s a breakdown of how I see the division:

In other words, we should delegate the computational. And double down on the relational.

​

The Agentic Framework in Practice

At Athletica, we don’t just run queries. We deploy agents: AI systems that make decisions, call other tools, and act in pursuit of an outcome.

Here’s what a real-world flow looks like:

1. Athlete uploads a training file or writes a journal entry.
2. LLM reviews the content, flags anomalies, and suggests a follow-up.
3. AI-agent determines whether a re-test is warranted or a load adjustment is needed.
4. Human coach reviews, adds narrative, adjusts as needed.

It’s still coaching. It’s just turbocharged.

The Future Coach’s Skill Stack

The modern coach will need to be more than just a good communicator or a physiology nerd. The next generation must:

• Read and interpret AI output
• Understand model limitations
• Know when to trust the machine and when to override
• Speak the language of data
• Ethically safeguard athlete autonomy

These are human skills. And they will matter more, not less, as AI grows in capability.

Final Thoughts: The Science 3.0 Era is Here

We’re moving fast now. LLMs are no longer academic curiosities. They’re practical assistants. They’re cognitive exoskeletons.

They won’t replace coaches. But they will replace coaches who refuse to adapt.

The real question: What parts of your coaching role would you gladly hand to an AI today?

Hit reply and let me know. I’ll share the responses in a future note.

​

Further Reading:
​

• ​The Computational Paths of Knowledge in AI Coaching
• ​HIIT Science book​
• ​Andreas Boecherer​

Paul Laursen, PhD

​​​​​​