Automatic Agent Optimization and the Rise of Self-Improving Agents

Writing prompts is dead! Prompt engineering was a fun phase, but it's over. As large language models evolve into multi-step agents with tools, memory, and retrieval, the real bottleneck isn't creativity it's context management. What agents see, remember, and decide now defines performance. Manual "prompt surgery" can't keep up. In this talk, I introduce automatic agent optimization a new class of algorithms that let agents rewrite themselves. Using the Opik Agent Optimizer SDK, we'll compare optimizers from GEPA to our new Hierarchical Adaptive Prompt Optimizer (HAPO), showing how they dynamically balance cost, latency, and accuracy. You'll see why the next generation of agentic AI will learn from its own mistakes, and why the best teams won't be writing prompts they'll be designing optimizers and curating datasets. Find out how leading AI research teams are building the future of agentic systems