Six weeks of relentless AI agent hype ran into a wall of engineering reality at two Silicon Valley events this week. At the Generative AI and Agentic AI Summit in San Jose and a separate AI event in Mountain View, technical staff from Google, Amazon, Microsoft, and Meta acknowledged that building and operating production AI agents remains expensive, unreliable, and operationally complex, CNBC reported on Saturday.
Token Economics Are the Hidden Cost Crisis
Kevin McGrath, CEO of AI startup Meibel, identified the core problem: companies are routing everything through large language models without considering whether a task actually requires one. “Just give all of your tokens and all of your money to an AI Claw bot that will just waste millions and millions of tokens,” McGrath told the summit, arguing that organizations need to be far more deliberate about which tasks warrant agent-level processing.
Google software engineer Deep Shah reinforced this in a session focused on managing operational costs at scale. “If you think of a machine learning system or any multi-agent system, there are multiple challenges you will find when you try to deploy that system at scale,” Shah said, according to CNBC. “The first one is the inference cost.”
Multi-Agent Orchestration Called “Chaotic”
Synchtron CEO Ravi Bulusu went further, describing the interdependencies between data organization, technology platforms, workforce structure, and agent coordination as fundamentally unresolved. “No single dimension is solved in isolation and the interdependencies are what make this hard, in fact chaotic even,” Bulusu told attendees.
The observation cuts at the central promise of autonomous agents: that they can seamlessly coordinate across enterprise systems. In practice, according to summit participants, the coordination layer itself becomes the bottleneck.
OpenClaw Criticized as Enterprise-Unready
The sharpest critique came from ThinkingAI co-founder Chris Han at Thursday’s Mountain View event. Han, whose company recently rebranded from mobile game analytics firm ThinkingData into an AI agent management platform, called OpenClaw insufficient for business use.
“OpenClaw is a good tool for personal things, but definitely cannot reach the enterprise level,” Han told CNBC. “In terms of the enterprise level, you have to figure out a lot of things, your memory, how to manage your agents, teams, communications; there are a lot of things you have to figure out.”
ThinkingAI has partnered with MiniMax, the Hong Kong-listed Chinese AI lab that open-sources its models, to offer enterprise agent management. Han declined to comment on national security concerns around Chinese AI models but noted the platform also supports OpenAI and Google models.
The Hype Correction in Context
The summit comments land after weeks of bullish signals. NVIDIA CEO Jensen Huang told CNBC’s Jim Cramer in March that OpenClaw “is definitely the next ChatGPT.” Enterprise vendors from Salesforce to Microsoft have rushed agent products to market. Funding rounds for agent startups have exploded.
None of the summit speakers argued that AI agents will fail. The consensus was more precise: organizations deploying agents without rigorous cost controls, architectural discipline, and governance frameworks will burn money rather than save it. The gap between demo and production remains wide, and the companies selling the tooling to close it are only now emerging.