Field Note — Skill-Tree Overview

SYSTEMS NOTE // 07  ·  JULY 2026  ·  SGI RESEARCH

AGENTIC AI
ROADMAP 2026

Adapted from a roadmap originally compiled by Anastasiia Shapovalova

Eleven stages from first script to production agent. Each stage separates what's actually required from what's optional polish, and lists the tools people reach for at that layer.

Must do Optional Tools / Tech

Click a legend tag to spotlight that category across every stage

Prompting◆ Agent Basics◆ LLMs & APIs◆ Tool Use◆ Frameworks◆ Orchestration◆ Memory◆ RAG◆ Deployment◆ Monitoring◆ Security◆ Prompting◆ Agent Basics◆ LLMs & APIs◆ Tool Use◆ Frameworks◆ Orchestration◆ Memory◆ RAG◆ Deployment◆ Monitoring◆ Security◆

The Eleven Stages

Build Order, Not
Just A Feature List

PINNED
01
Programming & Prompting
Tools / Tech
PythonJavaScriptTypeScriptShell / BashGo
Must do
Scripting & automationAPI requests (HTTP/JSON)File handlingAsync programmingWeb scrapingData formats (JSON/YAML)Prompting conceptsPrompt engineeringContext managementChain-of-thought promptsMulti-agent promptsGoal-oriented prompting
Optional
Self-critique & retry loopsReflexion loopingTask planning promptsRole promptingPrompt evaluation & guardrails
PINNED
02
Basics of AI Agents
Must do
What are AI agents?Autonomous vs semi-autonomous agentsAgent architectures (ReAct, CAMEL, AutoGPT, A2A)Model Context Protocol (MCP)A2A protocol (agent-to-agent)Goal decompositionTask planning algorithmsDecision-making policiesAction planning loopsMulti-agent collaborationSelf-reflection / feedback loops
PINNED
03
LLMs & APIs
Tools / Tech
OpenAI GPT-5.5Claude Opus 4.8Gemini 3.5 FlashMistral Small 4Llama 4 family
Must do
API authenticationRate limitingToolformer / function callingTool invocation & output parsingPrompt chaining via APIs
PINNED
04
Tool Use & Integration
Must do
Tool use systemMemory integrationExternal API callingFile reader / writer toolsPython execution toolsSearch & retrieval tools
Optional
Calculator & code interpreterWeb browsing toolsComputer use (headless browsers)
PINNED
05
Agent Frameworks
Tools / Tech
LangChainAutoGenCrewAIFlowiseAgentOpsHaystackSemantic KernelSuperagentLlamaIndexOpenPipe
PINNED
06
Orchestration & Automation
Tools / Tech
n8nMakeZapierLangGraph
Must do
DAG managementEvent-based triggersGuardrails & validationsLooping & conditional workflows
PINNED
07
Memory Management
Must do
Short-term memoryLong-term memoryEpisodic memoryVector stores
Tools / Tech
PineconeWeaviateChromaFAISS
PINNED
08
Knowledge & RAG
Must do
RAG (retrieval-augmented generation)Embedding modelsCustom data loadersDocument indexing
Optional
Query refinementHybrid searchRe-ranking models
Tools / Tech
LangChain RAGLlamaIndex RAGColBERTHybridSearch
PINNED
09
Deployment
Must do
API deploymentServerless functions
Tools / Tech
FastAPI / StreamlitDockerKubernetesVector DB hostingAgent hosting (Replicate, Modal, Vercel, Fly.io)
PINNED
10
Monitoring & Evaluation
Must do
Agent evaluation metricsHuman-in-the-loop feedbackCustom dashboards
Tools / Tech
LangSmithLogging / tracingOpenTelemetryPrometheusGrafana
PINNED
11
Security & Governance
Must do
Prompt injection protectionAPI key managementUser authenticationRole-based access control (RBAC)Output filteringData privacy & compliance (GDPR, AI Act)
Optional
Red team testing

THE STACK IS NOT
THE HARD PART.
KNOWING WHAT ORDER
TO LEARN IT IN IS.

More Roadmaps.
More Notes.
Source Roadmap content adapted from Anastasiia Shapovalova