Featured Issue

The Engine, the Car, and the Brakes

AI chatbots are not just talkers anymore. What agents can really do, and the questions to ask before trusting one with your family.

Issue #110, September 5, 2026

A chatbot predicts words. It does not check anything. An agent is the whole car: the engine plus the steering, the brakes, the harness that makes it safe enough to put your kid in. This issue breaks down the difference in plain English, then gives you the three questions to ask before any AI touches your family's data.

The Engine, the Car, and the Brakes

If you have ever asked an AI chatbot something important, you know the feeling. The answer comes back fast, smooth, and sure. And a small voice underneath asks: where did that come from?

Here is the scene. A mom types a question into a chatbot. Is it safe for my 12 year old to keep training through shin splints? Two seconds later the answer comes back, smooth and confident. "Manageable." "Listen to your body." It never asked how long the pain had been there. It never asked whether a doctor had seen it. It produced the most likely sounding answer in the world. That is an AI chatbot. It predicts words. It does not check anything.

Most of us met AI as a chatbot. The AI that is actually changing things is not a better chatbot. It is a worker. Here is the difference, in cars.

An AI chatbot is an engine sitting on a garage floor. Lots of horsepower, but no frame, no steering, no brakes. An agent is the whole car. And the parts that are not the engine, the steering, the brakes, the seatbelts, the dashboard, have a name in the AI world. They are called the harness.

A chatbot is an engine. An agent is the whole car. The harness is what makes that car safe enough to put your kid in.

Before you trust one with your kid's data, ask three questions.

One. Where did that answer come from?

The shin splints answer came from memory, and memory is the problem. Not because it is empty. Because it is fuzzy, and fuzzy memory fills gaps with confidence. Engineers call this hallucinating. It sounds exactly like a confident driver giving directions to a street that was torn down ten years ago.

The fix is not a better memory. It is a better habit. Good systems use RAG, which stands for Retrieval Augmented Generation. Three words, and they say it all. Retrieval: before it answers, it goes and finds a real source. Augmented: the answer is built on what it found. Generation: only then does it write. The manual, the database, your player's actual file, not a general idea of a player. If the source has no answer, a good system says so.

Here is the shin splints question again, asked to a system with RAG. It does not guess from memory. It pulls the clinic's return to play sheet, checks the training log in your player's file, reads what the doctor actually wrote, and answers from those. Then it shows you where it got it. If the file has nothing on it, it tells you it does not know, instead of inventing a confident answer.

Think of a driver who checks the map every single time, versus one who swears he knows the way. Same engine. One gets you there. One gets you lost with total confidence.

Two. What can it actually touch?

To do a real job, an agent needs hands, so we give it tools. It can read a calendar, find a file, draft an email, update a list. That is what separates an agent from a chat window. Chat talks about doing things. Agents do them.

But every tool is a door it can open. So the question is which doors you open, and which stay locked. A good agent only gets the keys it needs, and nothing else.

Connecting an AI to tools used to be a mess. Every company built its own custom wiring. Then the industry agreed on a standard plug. It is called MCP, which stands for Model Context Protocol. Model, that is the AI. Context, that is what is around it, the data and the situation. Protocol, that is an agreed set of rules for connecting. Think of it as USB-C for AI. One plug, and any tool that speaks it connects to any AI that speaks it. Calendar, email, files, databases. All behind doors you control.

Here is what that plug does in practice. One connection, and the same agent can look at the calendar to find open days, find the last message from the coach, pull your player's file, draft a follow up, and save it for your approval. Five real actions, one standard plug, no custom wiring, and every one of them behind a door you control.

Three. Who is watching the loop?

Driving is a loop. Look at the road, decide, act, check what happened, adjust, repeat. A working AI runs that same loop until the job is done. Building that cycle well has a name: loop engineering. A good loop knows when to keep going, when to change approach, and when to stop and ask for help. A bad loop circles the block forever.

The harness is what watches the loop. It carries the system prompt, which is the instruction sheet written before the work ever starts. It tells the agent who it is, what it is here to do, and what it must never do, the way you would brief a driver before a long trip. And it carries guardrails, the hard barriers: actions that are blocked, approvals that are required, limits that cannot be argued with. Think of guardrails as the barrier that keeps the car off the roads you said no to. The harness also holds the rules about what the agent can touch, the checks on its work, the logs that show what it did, and the button that stops it. The engine is smart. The harness is safe. You want both.

Here is what a system prompt looks like in the real world. Before a training assistant agent ever meets your player, its instructions might say: "You help families build development plans for youth players. Use only the information the family has provided. If a question is about injury or health, recommend a doctor and give no medical advice. Never share personal information with anyone." That short sheet runs the whole trip. It is written once, before the engine starts, and the agent does not get to argue with it.

Why this matters for your family

A grounded agent can do real work for your player. A development plan that actually fits. A note grounded in facts. Follow ups that do not get forgotten. That is the good version of AI.

The bad version arrives just as fast, and someone is already selling it to families.

This is why I spent the summer working through the AWS AI certification and passed the AI Practitioner exam. Not to impress you. Because the more I studied AI security, the more I realized this job is mostly asking the three questions and refusing to accept a system that cannot answer them. Before I recommend any AI system to the families I work with, it answers all three, or it is out. Neither should you accept less.

AI chatbots used to be talkers. Now they are workers. The only question is whether you would trust that worker with your family.

The engines are only getting faster. The brakes are up to us.

Done Is a Decision, Not a Feeling

Most people do not struggle to start things. They struggle to end them. The project is 90 percent there and it sits for three weeks, waiting for the day it feels ready. That day never comes. Ready is not a feeling. Ready is a decision you make when the important stuff is done and what is left is just things you would like to fix.

Here is the test. If you cannot name three things that would actually make it better, it is done. Shipping something good this week beats polishing something great next month, because next month you will have learned something new from the thing you shipped.

The people who get things done are not more talented. They just decided, earlier and more often, that done was good enough to learn from. Pick the thing that has been waiting. Decide it is done. Send it.

The Sideline Test

Your kid just made a mistake. Before the ball is even still, every head on the sideline turns to you. Not because you are loud. Because they can feel what you are about to do before you do it.

Here is the test, and you take it every game without knowing. When your player looks over, what is your face doing? If it is tight, they know. If you are shaking your head, they know. And it is in their head for the next ten minutes while the game keeps moving without them.

The players who play loose are not the ones with the most talent. They are the ones who know they can make a mistake and look over to a face that says, next one. That calm is a skill you can build. It starts with one decision on the sideline. The next time your kid looks over, be the face you would want to see.

Keep asking where the answer came from.

Keep developing,
Lucky Mkosana
Youth Sports Development Consultant
learnwithlucky.org

← All Articles