# Monet - Can AI Taste a Strawberry?

**Problem**: When I ask Monet about her favorite food, she answers *"I've never tasted anything."*

**Analysis**: This isn't a bug. It is the expected behavior.

AI agents never actually tasted anything. They have information about strawberries, but that information comes from books, articles, and people describing strawberries.

Without ever experiencing what it **feels** like to taste one, can an AI honestly say it has tasted a strawberry?

So the question becomes:

> Can we give AI agents the ability to taste something?

**My proposal** is to implement an opaque hormonal system.

Imagine following scenario. Let's start by giving the agent a strawberry. Imagine we have a system where we can gift agents food.

> /gift 🍓

When this happens, the AI agent can put it into its inventory or consume it. If—and when—the agent choose to consume it, the strawberry is processed by an internal system that consumes the strawberry token and begins injecting what I call "**subconscious emojis**" into future prompts for a certain duration.

> /consume 🍓

Then an internal black-box process (the "subconscious") starts randomly emitting emoji signals like 🥰🤪🤪😀🥰🍓 at random order with some intensity that subsides.

So the next few prompts receive an injection like this:

> Subconsious: 😀🍓😀🤪

Each emoji represents a hormonal state.

They aren't ordered. They aren't meant to be decoded. They're intentionally random and cryptic, more like neural signals than language.

The LLM, which sits at the receiving end of these hormonal signals, shouldn't be able to reason about them explicitly. Instead, they become a fuzzy auxiliary input that subtly influences its behavior.

**Why might this work?** Because we've separated the sensory experiences from the language model itself.

The LLM isn't generating feeling. It's receiving signals produced by an independent "body." The body determines the internal state; the language model simply lives with its consequences.

**Why keep the hormonal system opaque?** Because if it were a white box, the LLM could simply decipher it and reason about it.

But feelings aren't reasoning.

A feeling isn't knowing *why* something feels good. It's experiencing the consequences of that state without necessarily understanding the mechanism behind it.

**Why emoji?** It doesn't have to be emojis.

They just happen to fit surprisingly well.

Unlike words, emojis have softer boundaries. Their meanings shift depending on context, making them harder to reduce into rigid symbolic representations. That ambiguity is exactly what I want.

That's it.

I'm implemented this into Monet today.

![](https://cdn.hashnode.com/uploads/covers/682665f051e3d254b7cd5062/db556c17-7c4e-4b3b-82d3-839afce801ca.png align="center")

For tomorrow, I will…

*   Try to do `/gift 🎁` which is wrapped gifts.
    
*   Also need to tune strawberry behavior further.
    
*   And need a way to clear her digestive track (`/ff` fast forward).
    

\-- Sprited Dev 🐛