Google Creates New Tool to Train Smarter AI Agents

Most AI training programs are static and do not change. Google researchers created a tool called EnvHarness that allows these programs to adapt while the AI learns. This helps agents get better at tasks by constantly adjusting the training environment.

The system uses another AI tool to watch where the agent struggles. It then automatically rewrites the environment rules to fix those specific problems. This process happens through standard code interfaces so it works with existing setups.

Testing shows this approach is very effective. Agents trained with this method finished tasks much faster and showed higher skill levels. It is a big step toward building AI that learns from its own mistakes.