The Agent Hype Will Hit the Same Wall as RPA

Jensen Huang told Citadel Securities last week that Nvidia now has more AI agents than humans working in cybersecurity. Marc Benioff claims current CEOs will be the last to manage all-human workforces. Anthropic CEO Dario Amodei predicts AI could eliminate 50% of entry-level white-collar jobs within five years. Everyone's panicking about AI agents taking over the workforce.

I've seen this movie before.

Back in 2017-2018, we had the exact same panic about Robotic Process Automation. Same breathless predictions. Same "transform or die" urgency. The RPA market hit $846 million in 2018, with Deloitte predicting that 72% of organizations would adopt it within two years. Companies poured billions into UiPath and Automation Anywhere, convinced they were about to automate everything.

Then reality hit. Research found that 30-50% of initial RPA implementations failed. By 2022, RPA market growth had slowed from 63% to just 22%. The problem was simple: RPA automated the easy stuff brilliantly, but when processes required judgment, context, or handling exceptions, which is the stuff that makes work actually valuable, the bots fell apart.

Now look at what's happening with AI agents. KPMG reports that deployment has tripled since Q4, with 82% of business leaders believing agents will become valuable within 12 months. But when you dig into what's actually happening, you see something interesting.

When Huang says "100% of Nvidia engineers use Cursor AI", he's describing tools that assist humans, not autonomous agents that replace them. There's a massive difference between a tool that helps you write code faster and a bot that writes code independently while you sleep.

Even Benioff, the biggest AI agent evangelist in tech, admits the technology is "not 100% accurate" and that "you need the human in the loop." And when you look at the actual numbers from Salesforce, they cut 4,000 customer service roles, but agents now handle only 50% of interactions. The other 50% still requires humans. Why? Because half the work involves judgment calls, upset customers, edge cases, and situations the bot can't handle.

Anthropic tracks that 60% of AI usage is for augmentation and 40% for automation, and that split reveals the fundamental constraint. Automation works for well-defined tasks. Everything else still needs humans.

When Amodei warns that "AI is starting to get better than humans at almost all intellectual tasks", pay attention to the phrase "almost all." AI handles repetitive, rule-based, high-volume, low-ambiguity tasks exceptionally well. That's exactly what RPA promised to automate a decade ago, and it did…for about 20-30% of work before hitting the complexity wall.

The dangerous assumption everyone's making is that current AI agents can handle the remaining 70-80% of work that requires novel problem-solving, political navigation, judgment under uncertainty, adapting to constantly changing requirements, and understanding unstated context. That's the work that actually matters, and it's the same work that broke RPA implementations.

Here's what I expect will happen. Right now, companies are rushing to deploy agents, early success stories are proliferating, and consultants are getting rich selling AI transformation. Within 12-18 months, organizations will discover what RPA users learned: agents excel at narrow tasks but struggle with ambiguity. Within 24 months, the market will realize AI agents are powerful tools, not human replacements.

We're already seeing this play out. Salesforce is "radically reshaping" by redeploying thousands of employees rather than eliminating them: 51% of their Q1 hiring was internal. They're moving people around, not getting rid of them.

And notice what Benioff said about hiring: Salesforce won't hire engineers, support staff, or lawyers this year, but they are hiring salespeople and customer success employees. Why? Because building and deploying AI agents effectively requires human judgment, relationship management, and adaptation. The hard work hasn't gone away. It's just shifted.

Most companies will chase agent deployment numbers. A few will figure out something more valuable: which work should stay human. That second group will quietly pull ahead while everyone else drowns in coordination costs and broken automation.

So here's the contrarian question: What if the real competitive advantage has nothing to do with having more AI agents, and everything to do with knowing which work should stay human?

If you found this post helpful, consider sharing it with another executive grappling with AI, technology, and data. If you want to explore AI and other Technology strategies, grab some time on my calendar, and let's chat.

Startups who read beehiiv can receive a 90% discount on Intercom's AI-first customer service platform, plus Fin—the #1 AI agent for customer service—free for a full year.

That's like having a full-time human support agent at no cost.

What’s included?

Who’s eligible?

Intercom’s program is for high-growth, high-potential companies that are:

Apply now