The agent economy just hit its first real friction point: you can't delegate to something that needs more managing than an intern.

The Summary

The Signal

Rashidi ran four AI agents simultaneously. Two got fired. Not because they failed catastrophically, but because the supervision overhead exceeded the output value. "I don't have the time to babysit agents and keep course correcting the context," she said. This isn't a story about technology failing. It's about the hidden cost structure of delegation to machines that don't actually understand what you're asking.

The term "botsitting" comes from a recent Glean report that measured what everyone building with agents already knew but didn't want to quantify: maintaining AI agents is work. Real work. 6.4 hours per week of it, on average. That's debugging prompt drift, re-feeding context after the agent loses thread, and manually reviewing output to catch the confident-but-wrong answers that slip through.

"I was spending more time babysitting them than doing useful work."

Here's what makes this data point sharp: Rashidi isn't some skeptical executive dipping a toe into AI. She's an AI strategist. Her job is to understand and deploy these tools. MIT Tech Review notes that companies are framing AI tools as "coworkers" with names like Alex, creating an expectation of peer-level collaboration that the technology can't deliver yet. The metaphor breaks when you realize no human coworker requires you to spend 6.4 hours a week explaining basic context they should already have.

The replacement strategy matters here. Rashidi didn't abandon delegation entirely. She replaced the agents with human virtual assistants. The calculus: humans cost more per hour but require dramatically less supervision overhead. A human VA absorbs context once and retains it. An agent needs re-grounding every session. When you add up total cost of ownership, the human starts looking efficient again.

Key comparison:

  • Agent: Low hourly cost, high supervision overhead, context amnesia
  • Human VA: Higher hourly cost, low supervision overhead, context retention
  • Winner: Depends on task complexity and how much your time is worth

This creates a wedge in the agent market. Simple, repetitive tasks with zero context requirements still work fine. Scheduling, basic data entry, template generation. The moment you need judgment, memory, or the ability to course-correct without guidance, current agents fall apart. Companies are still figuring out the integration model, which is code for "nobody knows how to make this work at scale yet."

The Implication

If AI strategists are firing their agents, expect broader pullback in the "AI coworker" category over the next six months. Not abandonment, but right-sizing. The use cases that survive will be narrow, repetitive, and low-context. Everything else shifts back to humans or waits for the next model generation that actually retains context across sessions.

For anyone building agent products: the supervision overhead is your actual competitor, not other agent tools. If your agent saves someone 5 hours but costs 6 hours to manage, you've built a time-loss product. Measure success by net time saved after supervision costs. Anything less is marketing.

Sources

Business Insider Tech | MIT Tech Review AI