Silicon Valley AI agents face cost and coordination problems.

At two Silicon Valley events, executives and engineers said AI agents can automate office work but the underlying systems remain fragile; speakers warned that routing too much through large language models can waste tokens and raise inference costs.

Two events in Silicon Valley this week brought executives and engineers together to discuss the state of AI agents. Presenters described broad enthusiasm for agents that can automate office tasks and also raised concerns about fragile underlying systems. Several speakers warned that treating every task as a large language model problem can drive up usage and operating costs. What was reported: - Kevin McGrath, CEO of Meibel, said the biggest problem is the idea that everything must be processed by an LLM and warned that an "AI Claw" setup could waste millions of tokens and money. - The recent rise of OpenClaw, described as a harness for using multiple AI models to build and manage fleets of digital assistants, has intensified interest in agent systems. - Google software engineer Deep Shah highlighted operational challenges when deploying multi-agent systems at scale, saying inference cost is a primary concern and that poor monitoring can increase expenses. Summary: Speakers said AI agents can boost automation but also create cost and coordination challenges for organizations. Undetermined at this time.