OA
FreeAdvanced16 lessons2h 24m

Autonomous Self-Improving Agents

Build agents that learn your workflow, improve themselves overnight, and compound knowledge through Obsidian. Covers Karpathy autoresearch, GEPA evolutionary optimization, LLM Wiki memory, and the architecture of agents that get better every day without human intervention.

Course Content

Module 1: The Self-Improving Agent Thesis
19 min

Why agents that learn from their own work are the next frontier, and how autoresearch + GEPA + Obsidian create a compound improvement loop.

  • Why Static Agents Are Already Obsolete8 min
  • The Compound Loop: Work, Learn, Improve, Repeat8 min
  • Module Quiz+25 XP
Module 2: Karpathy's Autoresearch Method
19 min

Autonomous overnight experimentation: the three-file architecture, time-budgeted iterations, and how to apply it beyond ML training.

  • Autoresearch Architecture and Workflow8 min
  • Applying Autoresearch to Any Agent Task8 min
  • Module Quiz+25 XP
Module 3: GEPA: Evolutionary Agent Optimization
20 min

Genetic-Pareto optimization that reads execution traces, diagnoses failures, and evolves better agents automatically.

  • How GEPA Works: Generate, Evaluate, Reflect, Select9 min
  • GEPA Adapters: Prompts, DSPy, RAG, and MCP8 min
  • Module Quiz+25 XP
Module 4: Obsidian as Agent Memory
18 min

Your agent's persistent brain: every task enriches the wiki, every wiki page sharpens future performance.

  • Wiring Agents to Read and Write Obsidian8 min
  • The Knowledge Flywheel: Tasks That Make Agents Smarter7 min
  • Module Quiz+25 XP
Module 5: Building the Overnight Improvement Pipeline
18 min

Combine autoresearch + GEPA + Obsidian into a pipeline that makes your agents measurably better every morning.

  • The Nightly Pipeline Architecture8 min
  • Measuring Improvement: Baselines, Evals, and Drift Detection7 min
  • Module Quiz+25 XP
Module 6: Workflow-Aware Agents
19 min

Agents that learn YOUR specific workflow, not just generic tasks. Pattern recognition, preference learning, and context accumulation.

  • Teaching Agents Your Preferences and Patterns8 min
  • From Assistant to Partner: The Autonomy Gradient8 min
  • Module Quiz+25 XP
Module 7: Capstone: Build Your Self-Improving Agent
19 min

Wire up the full stack: Obsidian wiki + autoresearch experiments + GEPA optimization + your real workflow.

  • Setting Up the Full Stack8 min
  • The 30-Day Improvement Test8 min
  • Module Quiz+25 XP
Module 8: Bonus: Autoresearch for Claude Code Skills
12 min

Use the autoresearch methodology to take your Claude Code skills from 70% reliability to 97%+ with automated eval suites, iterative prompt mutation, and overnight optimization.

  • Why Skills Fail 30% of the Time4 min
  • Building Your Eval Suite and Running Autoresearch5 min
  • Module Quiz+25 XP

Buy Team a Coffee

Help keep the lessons free

1:1 Coaching

1-hour paid consult — $220