Foundations01
Direct response
Resolve a self-contained task in one model call.
A FIELD GUIDE TO AGENTIC SYSTEMS
Better agents.
By design.
Understand the flow. Choose the right approach.
Put it to work with a reusable prompt.
36approaches
72system prompts
5source collections
PATTERN IN FOCUS10 / 36
Orchestrator–workers
Delegate the work. Keep ownership.
CoordinationDynamic delegation
THE COLLECTION
Find your next approach.
From one well-formed response to a coordinated team. Start simple; add only the structure your task needs.
Foundations02
Single agent with tools
One agent owns the task and selects available tools.
Foundations03
Retrieval-augmented generation
Retrieve relevant evidence before generating a grounded answer.
Workflows04
Prompt chaining
Run fixed stages in order, passing each output forward.
Workflows05
Routing
Classify the request and choose its handling path.
Workflows06
Parallelization
Run independent work concurrently, then reconcile its results.
Workflows07
Bounded loop
Repeat a workflow until an explicit exit condition holds.
Workflows08
Iterative refinement
Improve a persistent artifact through successive bounded revisions.
Workflows09
Custom workflow
Compose rules, branches, agents, and checkpoints into one flow.
Coordination10
Orchestrator–workers
A central agent decomposes, delegates, and synthesizes the task.
Coordination11
Subagents as tools
A parent calls a specialist and receives its result.
Coordination12
Hierarchical delegation
Delegate recursively through a bounded tree of agent responsibilities.
Coordination13
Agent handoffs
Transfer active responsibility to another agent with relevant state.
Coordination14
Group discussion
Specialists collaborate in a shared, moderated conversation.
Coordination15
Structured debate
Opposing positions exchange evidence before an adjudicated conclusion.
Coordination16
Swarm collaboration
Peers coordinate directly without a central task-planning supervisor.
Coordination17
Adaptive orchestration
A manager revises a task ledger as evidence changes.
Coordination18
Role-based cooperation
Assign complementary responsibilities and explicit decision rights to agents.
Planning19
Goal clarification
Turn an expressed request into explicit, testable goals.
Planning20
Proactive goal proposal
Infer a useful proposed goal from permitted context.
Planning21
One-shot planning
Generate the complete proposed plan in one model invocation.
Planning22
Incremental planning
Generate the next plan segment using the latest observations.
Planning23
Single-path plan
Develop one ordered candidate path toward the goal.
Planning24
Multi-path planning
Explore alternative candidate paths before selecting one.
Planning25
Reason–act–observe
Choose an action, observe its outcome, then decide again.
Quality26
Self-reflection
Review and revise your own artifact against explicit criteria.
Quality27
Cross-reflection
Have another agent review an artifact and return feedback.
Quality28
Evaluator–optimizer
Generate, critique, and revise until acceptance or a limit.
Quality29
Voting & ensembles
Aggregate separate judgments using a predefined decision rule.
Quality30
Human-in-the-loop
Pause at a defined boundary for human input or approval.
Quality31
Guardrails
Constrain inputs, actions, and outputs with explicit checks.
Quality32
Agent evaluation
Test agent behavior against defined requirements and evidence.
Capabilities33
On-demand skills
Load specialized instructions while the same agent stays in control.
Capabilities34
Prompt & response adaptation
Normalize instructions or outputs while preserving their intended meaning.
Capabilities35
Tool & agent registry
Discover capabilities through a maintained catalogue of interfaces.
Capabilities36
Agent adapter
Translate between agent requests and an external system interface.
A GOOD PLACE TO START
Choose the problem.
Then the pattern.
Architecture, behavior, and infrastructure are different layers. The atlas keeps them distinct.