Workflow → Agent Evolution

Every agent pattern evolves from a workflow. When conditional logic becomes too complex to hardcode, the LLM takes over flow control.

Workflow

Evaluator-Optimizer

Generate-evaluate feedback loop that iteratively improves output.

Workflow

Orchestrator-Worker

LLM decomposes a task and delegates to specialized workers.

Workflow

Parallel Calls

Concurrent LLM calls on independent inputs, aggregated at the end.

Workflow

Prompt Chaining

Sequential LLM calls with validation gates between steps.

Agent

Agentic RAG

RAG where the agent plans retrievals, decomposes queries, routes across sources, reflects on sufficiency, and enforces citation-bound answers.

Agent

Event-Driven

Agents triggered by queue or stream events rather than HTTP requests.

Agent

Long-Horizon

Multi-session agent tasks that span hours to weeks; checkpoint-and-resume across crashes, deploys, and external waits.

Agent

Multi-Agent

Supervisor-worker delegation across multiple autonomous agents.

Agent

Plan & Execute

LLM creates a full plan upfront, then executes each step sequentially.

Agent

RAG

Retrieval-augmented generation: retrieve relevant context before generating.

Agent

ReAct

Reason-act loop: the LLM reasons, calls a tool, observes, and repeats until done.

Agent

Reflection

LLM critiques its own output and self-improves through structured feedback.

Agent

Routing

Intent classification dispatches inputs to specialized handlers.

Agent

Saga

Long-running, multi-step business processes that need compensation when an intermediate step fails.

Workflow

Agent Pattern

Hover to highlight · Click for details

Workflow	Evolves Into	What Gets Added
Evaluator-Optimizer	Reflection	Evaluator becomes the LLM itself via self-critique prompting
Orchestrator-Worker	Multi-Agent	—
Orchestrator-Worker	Plan & Execute	—
Parallel Calls	RAG	Vector search retrieval replaces static parallel splits
Parallel Calls	Routing	LLM intent classification replaces hardcoded branch logic
Prompt Chaining	ReAct	Dynamic tool selection + LLM-controlled looping replaces hardcoded steps
Prompt Chaining	Saga	—