Agentic AI

Agentic AI is an AI system that can has agency, hence the name. They can accomplish a specific goal with limited supervision. It consists of AI agents, which themselves use LLM/ML models that mimic human decision making. Agentic AI exhibits autonomous, goal-driven behaviour. Using this it can make decisions and act.

Most of the 'AI Software Engineering' notes are pretty broad, but are usually about LLM tooling, which are the new hotness.

The following is the 'Agent Loop' [[20260306123345-ai-custom-agents]] can be used to customise this loop, or [[20260306121850-ai-agent-skills]] can help teach new capabilities for a specific task.

It builds on [[20250813104808-generative-ai]] by using LLMs to function in dynamic environments. Rather than generating content, it applies generative outputs towards specific goals.

Agents can delegate tasks to subagents, helping optimise (read: lower) the amount of context that needs to be given to the main agent. Subagents work in a separate context and return only a summary. [[20260413123244-ai-subagents-orchestration]]

In terms of IDEs, every IDE has its own way of doing things as standards are evolving, e.g. [[20260306121823-ai-instructions]]

[[ai]]