Phase 15
Autonomous Systems
Agents that run without human intervention, safely.
Lessons (22)
- 01The Shift from Chatbots to Long-Horizon Agents
- 02STaR, V-STaR, Quiet-STaR — Self-Taught Reasoning
- 03AlphaEvolve — Evolutionary Coding Agents
- 04Darwin Godel Machine — Open-Ended Self-Modifying Agents
- 05AI Scientist v2 — Workshop-Level Autonomous Research
- 06Automated Alignment Research (Anthropic AAR)
- 07Recursive Self-Improvement — Capability vs Alignment
- 08Bounded Self-Improvement Designs
- 09The Autonomous Coding Agent Landscape (2026)
- 10Permission Modes for Autonomous Agents
- 11Browser Agents and Long-Horizon Web Tasks
- 12Long-Running Background Agents: Durable Execution
- 13Action Budgets, Iteration Caps, and Cost Governors
- 14Kill Switches, Circuit Breakers, and Canary Tokens
- 15Human-in-the-Loop: Propose-Then-Commit
- 16Checkpoints and Rollback
- 17Constitutional AI and Rule Overrides
- 18Llama Guard and Input/Output Classification
- 19Anthropic Responsible Scaling Policy v3.0
- 20OpenAI Preparedness Framework and DeepMind Frontier Safety Framework
- 21METR Time Horizons and External Capability Evaluation
- 22CAIS, CAISI, and Societal-Scale Risk