The AI Evolution
From Generative AI to Artificial General Intelligence — A Living Research Map
By Frank Riemer, AI Architect at Oracle
The 5 Waves of AI
Each wave builds on the last. Understanding where we are in this sequence is the difference between reacting to AI and architecting with it.
Foundation Models
Foundation Models
2020 – 2023- GPT-3 demonstrates few-shot learning at scale
- DALL-E and Stable Diffusion unlock image generation
- GPT-4 achieves multimodal reasoning
- LLaMA opens the open-source frontier
Impact: Proved that scale alone produces emergent capabilities. Established the transformer as the universal architecture for generative intelligence.
Generative AI Goes Mainstream
Generative AI Goes Mainstream
2023 – 2024- ChatGPT reaches 100M users in two months
- Claude introduces constitutional AI and long context
- Midjourney v5/v6 redefines visual fidelity
- Enterprise adoption accelerates across every industry
Impact: Shifted AI from research curiosity to daily tool. Every knowledge worker gains a reasoning copilot. The creator economy begins its transformation.
Agentic AI
Agentic AI
2024 – 2025- Claude Code ships agentic development workflows
- Cursor and Windsurf redefine IDE intelligence
- Model Context Protocol (MCP) standardizes tool connectivity
- Devin demonstrates autonomous software engineering
Impact: AI transitions from answering questions to completing tasks. Agents gain tools, memory, and the ability to operate across systems. The human role shifts from operator to architect.
Multi-Agent Systems
Multi-Agent Systems
2025 – 2026- Agent swarms coordinate on complex workflows
- Orchestration frameworks mature (LangGraph, CrewAI)
- Autonomous teams handle end-to-end business processes
- Human-in-the-loop governance becomes standard practice
Impact: Individual agents compose into systems. Organizations deploy agent teams that plan, execute, and self-correct. The personal AI Center of Excellence becomes viable for individuals.
AGI Horizon
AGI Horizon
2026+- Self-improving systems demonstrate recursive capability gains
- Alignment research intensifies as capabilities accelerate
- Consciousness and sentience debates enter mainstream discourse
- Regulation frameworks attempt to keep pace with progress
Impact: The frontier shifts from "what can AI do" to "what should AI do." Technical capability outpaces governance. The builders who understand both architecture and ethics define this era.
The Model Landscape
A structured view of the models shaping each modality. The landscape shifts monthly — this map tracks the architecturally significant players.
The Automation Stack
Models produce intelligence. The automation stack turns that intelligence into action. These four layers connect reasoning to real-world workflows.
n8n
Workflow OrchestrationVisual automation platform connecting 400+ services. Triggers, conditions, and branching logic without code. Self-hostable for full data sovereignty.
MCP
Tool Connectivity ProtocolModel Context Protocol standardizes how AI models connect to external tools, databases, and APIs. One protocol, universal compatibility across models and IDEs.
Claude Code
Agentic DevelopmentTerminal-native AI agent that reads, writes, and refactors entire codebases. Plans multi-step implementations, runs tests, and commits — with human oversight.
Vercel AI SDK
Streaming InferenceTypeScript SDK for building AI-powered applications. Streaming responses, structured outputs, and tool calling — production-ready from day one.
The pipeline: Vercel AI SDK streams model output. MCP connects models to tools. Claude Code orchestrates multi-step tasks. n8n automates the surrounding workflows. Together, they form a complete intelligence-to-action stack.
For Builders
Start building with AI today. These resources bridge the gap between understanding the landscape and shipping real systems.
Building enterprise AI systems or a personal AI stack?
This research map is maintained by Frank Riemer and updated as the landscape evolves. Last updated: March 2026.