Autonomous AI agents and reasoning models
Agentes de IA autónomos y modelos de razonamiento
- Date
- 2023-2025 · confidence medium
- Place
- OpenAI and Anthropic, San Francisco, United States; DeepSeek, Hangzhou, China (R1, 2025)
- Credited to
- Anthropic (computer use, October 2024; Claude Code, 2025); OpenAI (o1, 2024; Operator, 2025); DeepSeek (R1 with open weights, January 2025); the MCP protocol
- Field
- Knowledge & symbols
What it is
Systems that plan, use tools, browse, write and execute code and chain actions for hours with minimal supervision, supported by models that reason step by step before responding. They mark the shift from AI that answers to AI that acts.
A rapidly evolving field in 2025-2026; the dates of specific milestones vary depending on the system considered.
What it led to
- The automation of programming and office work, and the debate over employment
- The infrastructure race (gigawatt data centres, chips, nuclear power)
- The security and control risks of autonomous systems and international governance
Built on
- AlphaGo2016
- Transformer architecture2017
- ChatGPT and GPT-4
- reinforcement learning
Made possible
- agent-assisted software development
- automated scientific research
- AI tool-use protocols