A lot of forward-thinking leaders are running around right now telling people that if they don't have AI agents working for them, then they're falling behind.

But what's getting lost in the shuffle when it comes to agents is the phenomenon that Peter Steinberger, the founder of OpenClaw—the movement that jump-started the 2026 agent boom—has clearly talked about: the ways agents need humans.

"They are spiky smart, and they're really good at things, but if you don't navigate them well, if you don't have a vision of what you're going to build, it's still going to be slop," said Steinberger, in an interview with Peter Yang. "If you don't ask the right questions, it's still going to be slop."

I've been thinking about Steinberger's words a lot lately in the midst of all the current agent-mania. A recent study found that white-collar workers are facing an explosion of AI-generated "workslop" from chatbots spitting out documents with poor direction from humans—the same issue Steinberger highlighted. This is inundating workers with piles of these docs to sort out and clean up. As a result, 92% of executives say AI is making workers more productive, while 40% of workers claim it saves them no time at all.

Meanwhile, the buzz phrase that's been running rampant in the AI industry lately is "AI psychosis." This isn't the chatbot psychosis that refers to people who fall in love with chatbots or suffer a break from reality because of chatbot hallucinations. No, this type of AI psychosis was coined from a recent comment by AI pioneer Andrej Karpathy, and it's being referred to in the AI industry in near-heroic terms.

It's a type of token-maxing mania that AI coders experience when managing a swarm of agents, which they claim hugely boosts their productivity and causes them to work 18 hours a day, as they get hooked on constantly providing feedback to their agents and on how much they believe they can accomplish.

As I mentioned in my roundup from the HumanX conference, the people I spoke with in the AI industry at the event said the number of people running around claiming they are experiencing that kind of AI psychosis is greatly exaggerated, since it's being paraded as a badge of honor. Still, token-maxxing is being rewarded with little regard for the quality of the output.

Our Deeper View

In his April 16 TED Talk about how he created OpenClaw, Steinberger said that before OpenClaw he had been burned out and demotivated about creating software. But when he first tried coding agents in early 2025, he quickly discovered they automated "all the boring parts" of software development. "The bottleneck is no longer typing," he said. "It's thinking." Despite Steinberger's repeated emphasis on the importance of thinking and human direction of agents, the current agent mania has largely ignored this and blown through all wisdom and restraint. It would be smart for AI enthusiasts and enterprises to take note.