The AI agent revolution is happening. Just not the way anyone expected.

New research from A16Z, KPMG, and Writer reveals that agentic AI deployment has crossed 50% of enterprises. But here's the problem: companies are spending 93% of their AI budgets on tools and only 7% on the people who actually make them work.

The result? Most companies are getting zero measurable value from their AI agents.

**The Trust Problem**

Employee resistance is killing AI agent programs before they start. Despite massive corporate investments in AI tools, workers simply refuse to use them. The studies found that trust gaps between employees and AI systems are widening, not narrowing.

This explains why Anthropic just blocked OpenClaw users from flat-rate pricing. AI agents consume enormous compute resources, but most companies haven't figured out how to extract enough value to justify the costs.

**The 93/7 Problem**

Companies are making a classic mistake: buying technology without investing in the human systems needed to deploy it effectively. The 93/7 spending split between tools and people is backwards.

Successful AI implementations require training, change management, workflow redesign, and ongoing human oversight. But most companies are trying to solve human problems with more technology.

**The Security Bottleneck**

Forbes warned this week that "existing security frameworks are unprepared for the autonomous, cross-boundary nature" of AI agents. Enterprises are rushing to deploy systems they can't properly secure or monitor.

With Anthropic's Mythos showing how AI can exploit vulnerabilities at superhuman speed, companies deploying unmonitored AI agents are essentially handing attackers the keys to their infrastructure.

**What Actually Works**

The companies extracting value from AI agents are doing three things differently:

1. **Investing in people first**: Training employees, redesigning workflows, building trust through transparency

2. **Starting small**: Piloting agents in low-risk environments with clear success metrics

3. **Building security from day one**: Treating AI agents as potential attack vectors, not just productivity tools

**The Reality Check**

AI agents are not self-deploying productivity miracles. They're sophisticated tools that require sophisticated human systems to use effectively. The companies figuring this out first will capture most of the value. The rest will keep throwing money at technology that their employees refuse to use.