Automatic Agent Optimization and the Rise of Self-Improving Agents
summary
Introduces Opik Agent Optimizer SDK and two optimizer classes—an evolutionary/genetic optimizer (with population, mutation, hall of fame and Pareto scoring) and a Hierarchical Adaptive Prompt Optimizer (HAPO) that does root‑cause style failures → strategies → targeted mutations. Uses LangChain JSON mode and public HuggingFace datasets. Reports a concrete result: a JSON parsing task moves from 12% to 97% accuracy (≈700% improvement). Shows code, runs iterations, and publishes the tools and benchmarks.
AI-written summary · about
chapters (9)
- 0:00Introduction and setup
- 2:40Speed cost accuracy trade-offs
- 5:05From prompt to intent engineering
- 8:35Meta prompting and optimization
- 12:04Genetic and evolutionary optimizers
- 18:18Hierarchical adaptive prompt optimizer
- 22:51Real-world optimization results
- 23:32Future of prompt engineering
- 24:37Q&A session
related talks
01h 21mAgent Optimization with Pydantic AI: GEPA, Evals, Feedback Loops
14mUnconference: Self-optimizing deep research agents
45mAI agents in practice - beyond the prototype
35mThe Agentic AI Engineer
26mFrom autocomplete to agents: AI coding assistance state of play
07h 47mAI Agents Conference 2026 - Day 5
42mComputer-Use Agents: From SFT to Classic RL
34mAI Agents: Transforming Autonomous Decision-Making