I ask myself this question constantly from a developer’s perspective: Does the AI actually do the thing, or does it just tell my code to do the thing?
For a normal user, they can’t differentiate between an agentic AI and a generative AI. People tend to praise Agentic AI too much, forgetting that it just uses tools to handle the executions.
If we say "AI can book a plane for you," does it really do that? Or does it just lay out instructions to its tools to handle that?
Take this example: I created an application that leverages AI function calling capabilities. This app can do things automatically, such as notifying me if my laptop battery is below a certain level, managing OS functionalities like laptop shutdown, checking RAM information, and handling process management.
If I presented this app to a non-developer and said, "This app is AI-driven," they’d conclude that AI is very powerful and scary. They imagine some digital ghost living inside the machine making decisions.
But for me, I know I just used the function calling capability of the AI. I created these tools myself. The AI is just the orchestrator.
I always look at AI as just the brain. Without these tools, it’s more of an intelligent textbook. It has all the knowledge in the world, but it can’t turn the page, let alone shut down your Linux kernel.
The "magic" isn’t in the model’s autonomy. It’s in the engineering of the bridges we build between natural language and system execution. The AI doesn’t know how to read your RAM usage; it just knows that when you ask for it, it should trigger the get_ram_info() function I wrote.
Let’s stop giving the AI all the credit for "agentic" workflows. An agent is only as powerful as the tools a developer gives it. If you strip away the APIs, the bash scripts, and the function definitions, the agent reverts right back to being a chatbot that can’t touch the real world.
Next time you see a mind-blowing AI demo, look past the shiny UI. Appreciate the developer who mapped out the tool definitions, handled the error states, and wrote the actual execution logic.
The AI is the brain. We build the hands.
Top comments (1)
This "AI is the brain, we build the hands" framing is spot on. One thing I've been experimenting with: giving agents better "brain" controls through their own tool infrastructure. I built Brainstorm-Mode (mehmetcanfarsak/Brainstorm-Mode on GitHub) — a hook-based plugin that adds mode discipline to agents. Think of it like giving the brain a circuit breaker: PreToolUse hooks that block tool calls during ideation so the agent stays in thinking mode instead of jumping to execution. Three modes (divergent, actionable, academic) each enforce different constraints at the infrastructure level.