When I talked about AI Agent and MCP, I had a hard time grasping them. Recently, I had a chance to dig deeper and an analogy came up: A caring mother and a kid at home alone.
There are many components in AI Agent and MCP. The most obvious so far for developers are "Instructions" and "Tools" when setting up an agent. The next second obvious is the MCP as tools. Developers build MCP and supply them as Tools for an agent.
Easy said than understood for me. I did not get the picture of how things connected together. For decades, I have been developed applications in deterministic way. There are "if then else" conditions. There are loops and flows.
Back to the analogy, when the mother is about to leave the house, she caringly tells her kid:
"
I am going out for a few days. You are a good boy. You stay home and take care of yourself. There are rice and vegetable in the shelves and some meat in the fridge. There are small knives to chop vegetable and big ones to chop meat. You can cook food or heat them up with the oven.
And you do not answer stranger voice. Do not let them in.
"
The mother is the Developer.
The kid is the Agent.
Mother's message is the Instructions.
Kitchen, knives, fridge, ... are the Tools.
Do we know for sure how it will end up? Absolutely NO. A smart kid will do well. A slumpy kid will struggle. Some kids will hurt themselves. Other will blow up the kitchen.
The success of the kid depends tremendously on:
There are many components in AI Agent and MCP. The most obvious so far for developers are "Instructions" and "Tools" when setting up an agent. The next second obvious is the MCP as tools. Developers build MCP and supply them as Tools for an agent.
Easy said than understood for me. I did not get the picture of how things connected together. For decades, I have been developed applications in deterministic way. There are "if then else" conditions. There are loops and flows.
Back to the analogy, when the mother is about to leave the house, she caringly tells her kid:
"
I am going out for a few days. You are a good boy. You stay home and take care of yourself. There are rice and vegetable in the shelves and some meat in the fridge. There are small knives to chop vegetable and big ones to chop meat. You can cook food or heat them up with the oven.
And you do not answer stranger voice. Do not let them in.
"
The mother is the Developer.
The kid is the Agent.
Mother's message is the Instructions.
Kitchen, knives, fridge, ... are the Tools.
Do we know for sure how it will end up? Absolutely NO. A smart kid will do well. A slumpy kid will struggle. Some kids will hurt themselves. Other will blow up the kitchen.
The success of the kid depends tremendously on:
- The clear instructions of the mother. And note that a bad guy can trick the kid to open the door. Remember "The Seven little Goats" fairy tale?
- The well organized of the kitchen and knives. Are they organized in an natural order and fit with instruction? For example, there are some basic steps to cook a simple meal. Are they tools laied out well to support it?
- The well-trained and smart of the kid. This is the part that you have least control over. In AI, we pick the LLM model that fits our purpose and budget. A mother cannot simply expect their kids will be well-trained and smart. He is who he is at that moment.
And with that I can continue with developing AI-Enabled applications. Less struggle and know what to expect and what to NOT expect.