A structured reference of modern Agentic AI workflows, architectures, and orchestration patterns.
A single model completes the entire task using tools and a system prompt.
Best for simple workflows, prototypes, and low-complexity automation.
Multiple specialized agents cooperate by splitting a larger goal.
Useful for tasks requiring diverse skill sets, tools, or contexts.
Agents execute in a strict, linear order.
Ideal for predictable, step-by-step pipelines.
Multiple agents run simultaneously on the same input.
Speeds up processing but requires result resolution.
An agent repeats actions until a condition is met.
Great for monitoring, retries, and state-based workflows.
The agent alternates between reasoning (“thought”) and tools (“action”).
Enables multi-step navigation, planning, and adaptive behavior.
One agent generates outputs while another critiques them.
Ensures quality control for code, legal text, and risk-sensitive work.
Content is improved across multiple feedback cycles.
Works well for creative, complex, or polishing-heavy tasks.
A central manager decides which specialist agent should handle the task.
Flexible and adaptive for routing-based workflows.
A root agent breaks a large goal into subgoals handled by lower agents.
Strong for big, ambiguous, multi-stage objectives.
Agents collaborate as peers without a central controller.
Useful for brainstorming and divergent thinking, though costlier.
Execution pauses until a human approves the next step.
Critical for sensitive operations like deployments, refunds, or deletions.
