TrilokCloud · AI Design Patterns
AI design patterns, one day at a time
An ongoing series on the patterns behind reliable AI and agentic systems, written for builders who own the risk, not just the code.
ReAct: when your AI agent stops thinking and starts doing
Chain of Thought lets a model reason. ReAct lets it act on that reasoning. That shift changes everything about how failures happen and who owns them.
Reflection: the step that catches what reasoning misses
Chain of Thought reasons forward. ReAct grounds reasoning in real observations. Reflection turns the model around and asks it to look at its own output before anyone else does.
Tool Use: the moment your AI system crosses a boundary
The first three patterns work inside what the model already knows. Tool Use is the moment it reaches outside and touches real systems. That boundary crossing changes who owns the risk.
Plan and Execute: when an agent thinks before it acts
The first four patterns make a model smarter about individual steps. Plan and Execute is the first pattern that makes it think across steps before taking any of them.
Memory: what separates a capable agent from a useful one
An agent without memory starts from scratch every time. Memory is what allows it to carry context across steps, sessions, and tasks — and it is where some of the most underestimated production risks live.