Agent Architectures

Foundational Agent Types

AI agent architectures can be categorized based on their decision-making capabilities [5]:

Agent TypeDescriptionEnvironment
Simple ReflexActs based on current percept using condition-action rulesFully observable
Model-Based ReflexMaintains internal state to track aspects not evident in current perceptPartially observable
Goal-BasedUses goal information to choose among possibilitiesComplex decision-making
Utility-BasedUses utility function to map states to satisfaction levelsConflicting goals
LearningLearns from experiences to improve performance over timeUnknown environments

The ReAct Pattern

The ReAct (Reasoning and Acting) framework combines chain-of-thought reasoning with external tool use [6]. This iterative process enables agents to:

  1. Think: Reason about the current task
  2. Act: Decide on and execute an action (e.g., query a search engine)
  3. Observe: Process the outcome to inform the next step

Multi-Agent Architectures

Multi-agent systems (MAS) consist of multiple interacting agents that can be cooperative, competitive, or hybrid [7].

Common Orchestration Patterns

[8]:

PatternDescription
SequentialAgents executed in linear sequence
ParallelAgents executed concurrently
LoopAgent output fed back for iterative refinement
Review and CritiqueOne agent reviews another's work
CoordinatorCentral agent orchestrates others
Hierarchical Task DecompositionHigh-level agent breaks down tasks for lower-level agents
SwarmLarge number of agents work in parallel with minimal coordination

Deep Agents and Sub-Agents

Deep Agents (deepagents) is an open-source library from LangChain built on LangGraph that implements advanced agent patterns using filesystem-based context engineering [9]. The official repository is langchain-ai/deepagents.

Key concepts include:

  • Using the filesystem as a "scratch pad" to store and retrieve information
  • Storing long-term plans and findings of sub-agents
  • Enabling agents to "learn" by writing feedback and knowledge to files

Sub-agents are specialized agents that handle specific subtasks within a larger workflow, enabling:

  • Division of labor and specialization
  • Hierarchical task decomposition
  • Improved scalability and maintainability

Deep Agents Architecture

Deep Agents is described as "an agent harness built on langchain and langgraph. Deep Agents are equipped with a planning tool, a filesystem backend, and the ability to spawn subagents - making them well-equipped to handle complex agentic tasks."

Key features from the official repository:

  • Planning: Prior to task execution
  • Computer access: Giving the agent access to a shell and a filesystem
  • Sub-agent delegation: Isolated task execution

Available backends include:

BackendDescription
StateBackendEphemeral files stored in agent state (default)
FilesystemBackendReal disk operations under a root directory
StoreBackendPersistent storage using LangGraph Store
CompositeBackendRoute different paths to different backends

References

  1. GeeksforGeeks - Agents in AI
  2. IBM - What is a ReAct Agent?
  3. Microsoft Azure - AI Agent Design Patterns
  4. Google Cloud - Agentic AI Design Patterns
  5. LangChain - Filesystems for Context Engineering
  6. Deep Agents GitHub Repository