Jensen Huang said it in March: AI agents are “definitely the next ChatGPT.” Since then, the industry has gone all-in — OpenClaw, Claude Code Channels, hundreds of startups betting on autonomous agents. But at two events in Silicon Valley this week, the tone was different. Less euphoria, more sobriety.
Millions of Tokens for Nothing
Kevin McGrath, CEO of AI startup Meibel, put it bluntly: the biggest problem is the idea that everything needs to run through an LLM. “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.” His argument: companies need to be far more deliberate about which tasks actually need an AI agent — and which don’t.
The phenomenon even has a name now: “tokenmaxxing.” It describes the trend of pushing as many tokens through AI models as possible without asking whether it makes economic sense. For compute providers, it’s fantastic — for the companies footing the bill, not so much.
Interdependencies Make Everything Chaotic
The second problem goes deeper. AI agents don’t work in isolation — they access tools, call APIs, interact with each other. These interdependencies don’t just make systems complex, they make them “chaotic,” as speakers at the event described it. When one agent delegates a task to another, which then pulls external data, debugging becomes a nightmare.
This isn’t a theoretical problem. Companies report agents running in loops, hallucinating results, or simply executing the wrong actions — in production environments.
What This Means for Us
Does this mean AI agents are hype? No. But it means we’re still in the early days. The tools are getting better — Claude Code with its Routines, OpenAI’s Agents SDK with sandboxing, NanoClaw’s approval mechanism for sensitive actions. But the fundamental challenge remains: autonomous systems need clear boundaries, solid monitoring, and above all the honest question of whether an agent is truly the right approach for any given task.
The party isn’t over — but the hangover after the first night is real.
Sources: