← Knowledge Trees · Nestor G Pestelos Jr

AI Agents Knowledge Tree

A structured prerequisite curriculum and mastery map for autonomous AI agent architectures, loops, memory tiers, and multi-agent coordination.

Curriculum Status

22 Concepts organized across 6 learning phases · 14 Mastered · 6 Active Frontier · 2 Planned

Companion Reference: AI Agents Ground-Truth Reference · ELI5: How Do AI Agents Work?

Filter:
Phase 1: Agent Foundations & Decision Theory 3 nodes
Mastered AGT-01 Partially Observable MDP Formulation

Modeling agent trajectories as POMDPs $(\mathcal{S}, \mathcal{A}, \mathcal{O}, \mathcal{T}, \mathcal{R}, \gamma)$ where LLM token generation parameterizes a policy over discrete tool actions.

Prerequisites: Probability Theory, Autoregressive LLMs
Mastered AGT-02 Tool Calling Schemas (JSON / MCP)

Defining rigorous input/output schemas using JSON Schema and Model Context Protocol (MCP) to constrain and parse model tool invocations.

Prerequisites: Structured Data, JSON Schema
Mastered AGT-03 Observation & Feedback Grounding

Capturing runtime execution outputs (stdout, stderr, HTTP status, exit codes) and injecting them into the next turn prompt to ground reasoning in physical reality.

Prerequisites: POSIX Shell, Error Handling
Phase 2: Execution Loops & Reasoning Scaffolding 4 nodes
Mastered AGT-04 The ReAct Loop (Reason + Act)

Interleaving internal deduction with external tool calls (Thought → Action → Observation → Thought) to avoid speculative error compounding.

Prerequisites: Chain-of-Thought, Tool Calling
Mastered AGT-05 Plan-and-Solve Decomposition

Separating high-level strategic roadmap construction from tactical step-by-step execution to maintain long-horizon coherence.

Prerequisites: Task Decomposition, Directed Acyclic Graphs
Frontier AGT-06 Reflexion & Verbal Reinforcement

Generating post-mortem linguistic critiques on failed trajectories and persisting reflections into working memory to guide subsequent trials.

Prerequisites: Self-Critique, Working Memory
Frontier AGT-07 Tree-of-Thoughts & MCTS Planning

Exploring branching decision trajectories using Monte Carlo Tree Search or Breadth/Depth-first search with explicit state rollbacks and pruning heuristics.

Prerequisites: Search Algorithms, Value Heuristics
Phase 3: Memory Hierarchy & Context Management 4 nodes
Mastered AGT-08 Three-Tier Memory Architecture

Partitioning memory across L1 Working Context (active prompt), L2 Session Scratchpad (ephemeral logs), and L3 Long-term Storage (vector DB/vault).

Prerequisites: Context Engineering, Vector Databases
Mastered AGT-09 Context Truncation & Compaction

Preventing chat contamination and token window exhaustion by systematically summarizing completed milestones and dropping raw stdout logs.

Prerequisites: Token Budgeting, Progressive Summarization
Mastered AGT-10 Semantic Vector Retrieval (Episodic Memory)

Indexing past tool solutions, user preferences, and project facts into dense vector stores with cosine similarity $k$-NN lookups.

Prerequisites: Embeddings, RAG Architectures
Frontier AGT-11 Dynamic KV Cache Eviction & Caching

Structuring system instructions and tool definitions to maximize KV cache hit rates across multi-turn agent conversations.

Prerequisites: Attention KV-Cache, Prompt Caching
Phase 4: Action Execution & Sandboxing 4 nodes
Mastered AGT-12 Process Isolation & Containerization

Executing agent commands inside ephemeral Docker/WASM sandboxes with restricted network access and strict filesystem boundaries.

Prerequisites: Linux Namespaces, Containerization
Mastered AGT-13 Blast Radius & Destructive Command Gates

Intercepting high-risk system actions (destructive file deletions, migrations, terraform changes) with required human confirmation gates.

Prerequisites: Security Policies, Permission Hierarchies
Mastered AGT-14 Idempotent File Patching

Applying unified diffs or contiguous target-replacement blocks with exact collision detection to ensure deterministic codebase modifications.

Prerequisites: Unified Diff, Patch Mechanics
Frontier AGT-15 Transactional State Rollbacks

Automated checkpointing of filesystem and database states prior to multi-file refactors to guarantee clean one-command rollbacks on validation failures.

Prerequisites: Git Worktrees, SQLite Checkpoints
Phase 5: Multi-Agent Orchestration & Topologies 4 nodes
Mastered AGT-16 Hierarchical Supervisor-Worker Topology

An orchestrator agent decomposes tasks and dynamically spawns isolated worker subagents with scoped prompts and fresh context windows.

Prerequisites: Task Partitioning, Subagent Lifecycles
Mastered AGT-17 Sequential Pipeline Hand-offs

Structuring linear agent workflows where the validated artifact of one specialist becomes the input context for the next stage.

Prerequisites: Schema Contracts, Pipeline Verification
Frontier AGT-18 Blackboard Architecture

Coordinating asynchronous multi-agent fleets via a shared persistent store (git repository, SQLite DB) without point-to-point message passing.

Prerequisites: Distributed State, Optimistic Concurrency
Planned AGT-19 Multi-Agent Adversarial Debate & Consensus

Running parallel independent reasoning passes with structured cross-critique rounds to filter hallucinations and reach consensus.

Prerequisites: Game Theory, Consensus Mechanisms
Phase 6: Reliability & Defensive Guardrails 3 nodes
Mastered AGT-20 Anti-Loop Watchdogs & Circuit Breakers

Detecting repetitive error-fix cycles and halting execution after $N$ repeated failures to prevent infinite loops and token waste.

Prerequisites: State Fingerprinting, Circuit Breakers
Frontier AGT-21 Dual-LLM Privilege Separation

Isolating untrusted external data ingestion in a quarantined reader model before passing verified structured summaries to action agents.

Prerequisites: Prompt Injection Security, Privilege Separation
Planned AGT-22 Formal Verification of Agent Invariants

Mathematical specification and static assertion checking over agent plans prior to executing production-level mutations.

Prerequisites: Formal Methods, Temporal Logic