Google has open-sourced Harness-squared, a framework that lets artificial intelligence agents self-improve their prompts, tools, and memory setups without altering model weights. The system compares candidate runs to boost performance across tasks without needing an answer key.