A lot is said about AI agents and what it can do, here is the type of AI agents that we at the Stellarium Foundation mean, when we talk about AI agents.

Imagine you have a digital company comprised of only AI Agents, each agent is an AI that is trained differently and each agent has a different set of actions it can perform.

There are many kinds of AI today, there are general AI models like GPT-4o, Claude 3.5 Sonnet, Deepseek R1, Gemini 2.0. These models are trained in a huge and a wide amount of data and are very smart generally, but what we need are specialized models that can be deployed as AI agents. Let's say you want to create an AI agents lawyer, you would let these already smart AI models study on everything related to legal proceeding, all cases and decisions and laws and applications and everything, then as this AI “study” and adjust their weights to be a lawyer as a person would do (not just embedding or RAG) you would have an AI lawyer agent that could work digitally for you with superhuman performance. Then, for it to work with you, you would add actions, like send emails, access certain websites, customs actions, and so on and so forth, so that it could be an actual working entity. So that's how it would work.

What we need: A protocol or platform for AI agents, where you could “hire” these specialized AI agents, customize actions and add customized actions, and then these specialized AI agents could then be add into your digital company to be working entities, they could talk to each other and delegate tasks and all just like a person would in a corporation.

And this is why I have devised the Water Company

Water Company

To summarize how it works:

How AI Agents Are Supposed To Be Built

classDiagram
class AI_Agent {
+AI_Model_Executive
+AI_Model_Reasoner
+PersonalKnowledgeBase
+TeamKnowledgeBase
+Actions 
}
class PersonalKnowledgeBase{
+WorkDocuments+BestPractices
+JobInstructions
}
class Actions {        
+SendEmail()
+CheckEmail()     
+TakeScreenshot()
+PostOnSocial()
}

AI_Agent <|-- Actions
AI_Agent <|-- PersonalKnowledgeBase
classDiagram
class TeamKnowledgeBase{
+Benchmarks
+Metrics
+WorkPractices
}
class AI_Team {    
+AI_Agent_TeamManager        
+AI_Agent_Performance_Analyst
+AI_Agent_Media_Manager
+AI_Agent_Finance_Analyst
+AI_Agent_Marketing
+TeamKnowledgeBase
}
AI_Team <|--  TeamKnowledgeBase