Phase 15 · Autonomous Systems
Learn Autonomous AI Systems: 22 Free Lessons
Agents that run without human intervention, safely.
- 22 lessons
- ~20 hours
- Python
Start Phase 15
First lesson The Shift from Chatbots to Long-Horizon Agents
Run this command from the repository root:
python3 phases/15-autonomous-systems/01-long-horizon-agents/code/main.pyKeep the command, exit code, horizon projection, compounded reliability table, and one operational control you would require before a long run.
All 22 lessons in Phase 15
- The Shift from Chatbots to Long-Horizon Agents
In 2023 a chatbot answered a question in one turn. In 2026 a frontier model routinely runs minutes to hours on a single task.
- STaR, V-STaR, Quiet-STaR — Self-Taught Reasoning
The smallest possible self-improvement loop sits inside the rationale. A model generates a chain of thought, keeps the ones that land on correct answers, and fine-tunes on those. That is STaR.
- AlphaEvolve — Evolutionary Coding Agents
Pair a frontier coding model with an evolutionary loop and a machine-checkable evaluator. Let the loop run long enough. It discovers a 4x4 complex-matrix multiplication procedure that uses 48 scalar…
- Darwin Godel Machine — Open-Ended Self-Modifying Agents
Schmidhuber's 2003 Godel Machine required a formal proof that any self-modification was beneficial before accepting it. That proof is impossible in practice.
- AI Scientist v2 — Workshop-Level Autonomous Research
Sakana's AI Scientist v2 (Yamada et al., arXiv:2504.08066) runs the full research loop: hypothesis, code, experiments, figures, writeup, submission.
- Automated Alignment Research (Anthropic AAR)
Anthropic ran parallel teams of Claude Opus 4.6 Autonomous Alignment Researchers in independent sandboxes, coordinating via a shared forum whose logs live outside any sandbox (so agents cannot…
- Recursive Self-Improvement — Capability vs Alignment
Recursive self-improvement (RSI) is no longer speculation. The ICLR 2026 RSI Workshop in Rio (April 23-27) framed it as an engineering problem with concrete tooling.
- Bounded Self-Improvement Designs
Research has converged on four primitives for bounding a self-improvement loop. Formal invariants that must hold across every edit. Alignment anchors that cannot be modified.
- The Autonomous Coding Agent Landscape (2026)
SWE-bench Verified went from 4% to 80.9% in under three years. Same Claude Sonnet 4.5 scored 43.2% on SWE-agent v1 and 59.8% on Cline autonomous — the scaffolding around the model now matters as…
- Permission Modes for Autonomous Agents
A permission ladder — graduated levels of autonomy from review-every-action to approve-everything — is how a harness governs what an autonomous agent may do without asking.
- Browser Agents and Long-Horizon Web Tasks
ChatGPT agent (July 2025) merged Operator and deep research into one browser/terminal agent and set BrowseComp SOTA at 68.9%.
- Long-Running Background Agents: Durable Execution
Production long-horizon agents do not run in while True. Every LLM call becomes an activity with checkpoint, retry, and replay. Temporal's OpenAI Agents SDK integration went GA March 2026.
- Action Budgets, Iteration Caps, and Cost Governors
A mid-sized e-commerce agent's monthly LLM cost jumped from $1,200 to $4,800 after its team enabled the "order-tracking" skill. That is not a pricing bug.
- Kill Switches, Circuit Breakers, and Canary Tokens
A kill switch is a boolean held outside the agent's edit surface — a Redis key, a feature flag, a signed config — that disables the agent entirely.
- Human-in-the-Loop: Propose-Then-Commit
The 2026 consensus on HITL is specific. It is not "the agent asks, the user clicks Approve." It is propose-then-commit: the proposed action is persisted to a durable store with an idempotency key;…
- Checkpoints and Rollback
Every graph-state transition persists. When a worker crashes, its lease expires and another worker picks up at the latest checkpoint. Cloudflare Durable Objects hold state across hours or weeks.
- Constitutional AI and Rule Overrides
Anthropic's January 22, 2026 Claude Constitution runs 79 pages and is CC0. It moves from rule-based to reason-based alignment and establishes a four-tier priority hierarchy: (1) safety and…
- Llama Guard and Input/Output Classification
Llama Guard 3 (Meta, Llama-3.1-8B base, fine-tuned for content safety) classifies both LLM inputs and outputs against an MLCommons 13-hazard taxonomy across 8 languages.
- Anthropic Responsible Scaling Policy v3.0
RSP v3.0 went into effect February 24, 2026, replacing the 2023 policy. Two-tier mitigation: what Anthropic will do unilaterally vs what is framed as an industry-wide recommendation (including RAND…
- OpenAI Preparedness Framework and DeepMind Frontier Safety Framework
OpenAI Preparedness Framework v2 (April 2025) introduces Research Categories — Long-range Autonomy, Sandbagging, Autonomous Replication and Adaptation, Undermining Safeguards — distinct from Tracked…
- METR Time Horizons and External Capability Evaluation
METR (ex-ARC Evals) is an independent 501(c)(3) since December 2023. Their Time Horizon 1.1 benchmark (January 2026) fits a logistic curve to task-success probability vs log(expert human completion…
- CAIS, CAISI, and Societal-Scale Risk
The Center for AI Safety (CAIS, San Francisco, founded 2022 by Hendrycks and Zhang) publishes the four-risk framework — malicious use, AI races, organizational risks, rogue AIs — and the May 2023…
Glossary terms in this phase
- AgentA software system that lets a model select actions toward a goal, observe tool or environment results, and continue under an orchestration…
- AlignmentThe effort to make a model or AI system behave in ways that match intended goals, constraints, and human preferences across both expected…
- CheckpointA durable snapshot used to resume from a known boundary. In a workflow, it stores operational state and artifact references.
- Circuit BreakerA reliability control that temporarily stops calls to a dependency after failures cross a threshold, then probes whether the dependency…
- Coding AgentAn agent specialized for software work that can inspect a repository, edit files, run development tools, and use their outputs to advance…
- DPO (Direct Preference Optimization)A preference-optimization objective that trains a policy directly from preferred and rejected response pairs relative to a reference policy.
- Durable ExecutionRunning a workflow so its state and completed steps survive process crashes, restarts, or long waits without redoing confirmed side effects.
- Evaluation (Eval)A defined process for measuring model or system behavior on representative tasks using explicit success criteria, data, scorers, and…
- GuardrailsSystem controls that constrain inputs, tool use, outputs, permissions, and escalation.
- Human-in-the-Loop (HITL)A workflow design in which a person supplies judgment, correction, approval, or escalation at defined points in an AI-driven process.
- IdempotencyThe property that repeating the same operation with the same identity does not create additional side effects beyond the first successful…
- LLM (Large Language Model)A language model with enough capacity and broad training to perform many language tasks through prompting or adaptation.
- RollbackRestoring a previously known deployment or configuration when the current release violates operational, quality, or safety criteria.
- SaturationThe degree to which a constrained resource or service has exhausted its capacity, including queued work that cannot begin promptly.
- TokenAn integer identifier produced by a model-specific tokenizer from text, bytes, images, audio, or another input representation.
Frequently asked questions
How many lessons are in Phase 15: Autonomous Systems?
Phase 15 has 22 lessons, all Learn lessons. The lesson code uses Python.
What should I know before I start Phase 15?
The phase guide gives these prerequisites: Phase 14 Lesson 01, The Agent Loop. In the course roadmap, this phase builds on Phase 14: Agent Engineering.
Is Phase 15 free?
Yes. All 22 lessons are free to read on this site, and you do not need an account. The lesson code is open source under the MIT license.
How long does Phase 15 take?
The time estimates of all 22 lessons add up to about 20 hours.
What comes after Phase 15?
Phase 16: Multi-Agent & Swarms and Phase 18: Ethics, Safety & Alignment build on this phase.