Autonomous Agent History (BabyAGI → 2026)
From the AutoGPT / BabyAGI hype era to Claude Code / Operator — a 3-year maturation story.
March 2023 — AutoGPT: First viral autonomous agent. GPT-4 + self-prompting + web search + file I/O. Huge hype but often lost in loops, burned $50+/hour.
April 2023 — BabyAGI: Yossi Nakajima's 100-line prototype. Task creation/prioritization/execution loop. Pure concept, limited results.
End of 2023 — LangChain Agents: First popular framework. ZeroShotAgent, ReAct executor. Unstable but instructive.
2024 — Reasoning models (o1, Claude thinking): Reasoning now inside the model. Agent loops got smarter.
Late 2024 — MCP: Anthropic's tool standard. "USB-C for AI tools".
2025 — Claude Code, Operator, Computer Use: Agentic IDEs and browser agents went mainstream. 60%+ on SWE-Bench.
2026 — Multi-agent + long-running: CrewAI/AutoGen to production, durable workflow (Temporal/Restate) integration.