← Back to Podcast

AI in Sports

Episode 5, September 7, 2026

Your 12 year old comes in from practice wincing. The shins have hurt for a week. You type the question into a chatbot and two seconds later you get a smooth, confident answer. The answer came out of the model's memory. It never checked anything. Memory is the fuzziest part of the whole machine.

That is the gap this episode lives in. A chatbot predicts words. It does not verify them. And the tools have moved past talking. An agent can open files, run a schedule, and act on real data. Once software can do work on your player's file, the interesting question stops being whether it is smart. It becomes whether it is safe.

In this episode we break down:

• Why a confident answer is not the same thing as a checked answer, and why the difference matters more for a 12 year old than for an adult
• Retrieval augmented generation in plain English. The answer has to come from a real source, and a good system shows you the source
• MCP, the standard plug that decides which doors an agent can open. Every tool it gets is a door
• Loop engineering, system prompts, and guardrails. The engine is smart. The harness is what keeps you safe
• The one line that should stop a conversation dead. If the developer cannot answer all three questions in plain language, walk away

The three questions:

• Where did the answer come from?
• What can it actually touch?
• Who is watching the loop?

The engine, the car, and the brakes. You would not put your family in a car with no brakes just because the engine is impressive. The same standard applies to software that holds a young athlete's health notes and training logs.

Read the companion piece: The Engine, the Car, and the Brakes.

Subscribe on YouTube for more episodes.