AI Agents Just Built a Society. Can Your Org Even Approve a Pilot?
Why AI is not your bottleneck, but your leadership model is
Early this month, I watched thousands of AI agents build Moltbook, a social network where humans couldn’t even post. They debated, traded crypto, started a religion. It took them 48 hours. Your company’s been trying to get a chatbot pilot approved for six months. Do you see the problem?
While everyone else was laughing at robot religion, I was thinking about what this meant for my clients trying to get basic automation approved.
We just watched thousands of AI agents organize themselves into a working society. Meanwhile, inside the Fortune 500, most leaders are still trying to figure out how to get a chatbot to summarize a PDF correctly.
The problem isn’t that we lack tech. We lack leaders who understand what we’re building.
Thinking about Roles and not Tasks
The agents on Moltbook were interacting with one another. They had roles. They had a hierarchy. They had rules of engagement.
A big missing piece in AI strategies, if they exist at all, is still hiring prompt engineers when we should be hiring agent architects and cross-functional teams who are proficient in AI usage.
If you want to move beyond the pilot stage in 2026, you need a different type of thinking – designing org charts with AI agents in the middle of workflows.
Here is an example: You need one agent that figures out what needs to happen, another that actually does the work, and a third that checks if anything went wrong. That’s it. Three roles, not three thousand approval steps.
The False Choice between Controls and Nothing!
One of the frequent statements, often in regulated industries, I hear from leaders in my consulting work is “we want everything to be approved by people since I am worried about being sued for AI decisions.” In other words, 100% Human-in-the-Loop” (HITL). On the flipside, some leaders, especially in tech, want to go full autonomous on everything, like letting AI handle customer service end-to-end. Both extremes miss the point.
HITL is safe, but it doesn’t scale. It turns your expensive human talent into a bottleneck for cheap digital labor.
To lead a “Silicon Workforce” like the one we saw on Moltbook, you must graduate to Human-on-the-Loop (HOTL).
HITL: You read every email the agent drafts before it sends. (Slow, Low ROI). For example, a loan approval final decision should be done by a person.
HOTL: You set the policy (”Don’t offer a discount >15%”), and you monitor the dashboard of outcomes. You only intervene when the system throws an exception.
In our reconciliation software, the AI matches transactions automatically. If the confidence score is less than 90% about a match, it flags a human. That’s HOTL - AI does the work, humans handle exceptions. For fraud detection, it’s pure HITL - the system alerts the team immediately when it spots something suspicious. We use techniques like Isolation Forests and Local Outlier Factor models, which basically flag transactions that don’t match normal patterns.
Moltbook proved agents can coordinate at scale. So why don’t you trust yours to do the same?
The Leadership Challenge
Look, your job isn’t to be the smartest person anymore. It’s to set up the system so the agents can do their thing without you micromanaging. If you can’t let go of that control, you’re going to be the bottleneck, and not the technology.
Those Moltbook agents built a social network in 48 hours because nobody was standing in their way asking for approval at every step. The lesson isn’t to eliminate all controls, but to stop confusing oversight with obstruction.