Google Cloud has introduced Cloud Run instances, a new solution for running long-lived, stateful workloads like AI agents at a low cost. Traditional Cloud Run services scale to zero and are unsuitable for continuously running agents. Dedicated virtual machines, while always available, incur higher costs and require more management. Cloud Run instances offer dedicated, singleton compute runtimes that run a single instance without autoscaling. They support continuous runtime for up to seven days with a default automatic restart policy. Each instance is assigned a persistent HTTPS URL that remains stable across updates and restarts. Users can also stop and resume instances as needed. The cost for a continuously running instance for 30 days is as low as $5.70, utilizing shared vCPUs with burst budgets. This model is ideal for agents with intermittent compute-intensive tasks. Deploying AI agents like OpenClaw is simplified, with an easy command-line deployment after uploading configuration files. These deployed instances can run indefinitely and be interacted with via various communication platforms. Google Cloud users are already reporting positive results, with OffDeal noting significant reductions in cold starts and high reliability. Cloud Run instances are currently in preview and offer a cost-effective option for this new class of workloads.