One Billion AI Agents and Nobody's Driving

The real AI risk isn’t job loss. It’s organizational chaos.

Imagine walking into your office on a Monday morning and discovering that dozens of new employees had been hired over the weekend. Nobody interviewed them. Nobody checked their references. Nobody knows exactly what they’re working on or which files they have access to. But there they are, already busy at their desks, making decisions and taking actions on behalf of your company.

This is happening right now in enterprises around the world. Except the new employees aren’t people. They’re AI agents.

IDC projects that by 2029, there will be over one billion AI agents actively deployed across global enterprises. That’s a forty-fold increase from current levels. In the first half of 2025 alone, agent creation surged by 119 percent. We’re not talking about simple chatbots answering customer service questions. We’re talking about autonomous software that can execute business logic, access sensitive data, and take actions without a human in the loop.

And here’s the part that should keep executives up at night: most organizations have no idea how many agents they’ve already deployed, what those agents are doing, or what data they can touch.

I’ve been thinking about this problem for weeks now, and I’ve come to believe we’re having entirely the wrong conversation about AI risk. The headlines fixate on job losses and robot uprisings. But the more immediate danger is far more mundane and far more likely: organizational chaos. We’re adopting AI faster than we can govern it, and the consequences are already starting to show.

---

Let me tell you a story you’ve probably lived through, even if the details were different.

It’s 2015. Cloud computing has taken off. Your marketing team decided they needed a new analytics platform, so they signed up for one using a corporate credit card. Sales adopted their own CRM because the official one was too slow to get approved. The engineering team spun up AWS instances for a prototype that somehow became production infrastructure. Finance was running sensitive reports through a SaaS tool that IT had never heard of.

This was shadow IT. Business units moving faster than central governance could keep up. The result was a sprawling, ungovernable mess of overlapping tools, redundant costs, and security blind spots. It took years and millions of dollars to untangle.

We’re about to repeat the same mistake. The difference is that shadow IT involved passive tools that stored data and ran reports. Shadow AI involves active agents that make decisions and take actions. The tools we couldn’t see before just sat there. The tools we can’t see now actually do things.

Right now, somewhere in your organization, marketing has probably deployed an AI agent on one platform. Logistics built something on another. Finance is experimenting on a third. The data science team has homegrown tools running on internal servers. Each deployment made sense in isolation. Each team was trying to move fast and solve real problems. But nobody has a consolidated view of what exists, what it does, or what could go wrong.

MuleSoft recently expanded its Agent Fabric platform specifically to address this problem, introducing automated scanning tools that patrol major AI ecosystems looking for agents that nobody registered. Think about what that means. Enterprise software companies are now building tools whose primary purpose is to help CIOs find AI that’s already running inside their own organizations. The fact that there’s a market for “agent discovery” tells you everything about the current state of governance.

---

While enterprises grapple with this operational chaos, the public conversation about AI has been stuck on a single anxious question: will it take our jobs?

The worry is understandable. The International Monetary Fund estimates that over 60 percent of jobs in the developed world are “AI-exposed.” Anthropic’s CEO has floated the possibility of unemployment rates as high as 20 percent. Every week brings new headlines about layoffs attributed to AI efficiency gains.

But when I dug into what the research actually shows, a different picture emerged.

Recent studies indicate that only about 2.5 percent of jobs are currently at risk of full automation. And when researchers analyzed how people actually use AI tools in their work, they found something surprising: 60 percent of job-related queries were about augmentation, not automation. People aren’t asking AI to replace them. They’re asking AI to help them do their jobs better.

This pattern echoes every major technological disruption we’ve lived through. And every time, the predictions of mass unemployment turned out to be wrong.

Consider what happened when the steam engine transformed textile production in the early 1800s. The disruption was brutal and immediate. Hand-loom weavers watched their wages get cut in half within 15 years. Families who had practiced their craft for generations suddenly couldn’t feed themselves. If you were a weaver in 1810, the technology was genuinely destroying your livelihood.

But zoom out and the picture changes. Steam power made cloth dramatically cheaper. Demand exploded. Textile consumption soared among populations who previously couldn’t afford quality fabric. And the resistance of steam engines created entirely new categories of work that nobody anticipated. Coal mining expanded massively. Railroads needed maintenance crews. Cities grew, creating urban retail jobs that hadn’t existed before. The economy didn’t shrink. It transformed in ways the hand-loom weavers couldn’t have imagined.

The same story played out with electricity. When factories switched from steam to electric power in the early 1900s, certain jobs vanished almost overnight. Lamplighters. Icemen. Workers who maintained the complex belt-and-pulley systems that distributed mechanical power through factories. If your job was trimming gas lamp wicks, electricity was an extinction event.

Yet something remarkable happened as the cost of light, sound, and motion collapsed toward zero. Movie theaters became possible because projectors and indoor lighting were suddenly safe and affordable. Skyscrapers became practical because electric elevators could carry people thirty stories. Household appliances transformed domestic labor. Retail stores could stay open after sunset. Nobody watching Thomas Edison flip the switch at Pearl Street in 1882 could have predicted that electricity would create the cinema industry, reshape urban architecture, and change how families spent their evenings. The technology eliminated specific jobs while enabling entire new industries.

JP Morgan’s research team recently traced this pattern across centuries of technological change. They noted that the steam engine took 61 years to show up meaningfully in productivity statistics. Electricity took 32 years. Personal computers took 15. By their estimates, AI might take only seven.

The speed is accelerating. The destination remains the same: disruption and transformation, not destruction.

---

So if mass unemployment isn’t the real risk, what should we actually be worried about?

I keep coming back to the governance gap. We’re deploying AI faster than we understand what we’re deploying. And the consequences of that gap are both more subtle and more immediate than the job loss fears suggest.

Think about what happens when a company can’t answer basic questions about its own AI. How many agents are running? What tasks are they performing? What data can they access? How do those tools connect to actual business outcomes? Are we paying for redundant capabilities across different teams?

Most enterprises can’t answer any of these questions with confidence. They’re spending millions on AI while flying blind about what they’ve bought.

This is where the jobs conversation and the governance conversation collide. The fear that AI will replace human judgment isn’t wrong. But it’s not wrong because AI is too capable. It’s wrong because we’re deploying AI without exercising enough human judgment in the first place.

I use AI constantly in my own work. I can’t imagine going back to working without it. The leverage is real. But that leverage cuts both ways. AI is extraordinarily good at pattern recognition, logical consistency, and optimization at scale. Call it left-brain brilliance. It can process information faster than any human, spot patterns in data we’d never notice, and execute repetitive tasks without fatigue.

What AI cannot do is exercise the kind of judgment that makes those capabilities valuable. Common sense about novel situations. The ability to infer cause from correlation. Emotional intelligence that reads context and nuance. And perhaps most importantly, the wisdom to know what’s worth doing in the first place.

The organizations that thrive in this era won’t be the ones that automate the most. They’ll be the ones that draw the clearest lines between what AI should handle and what humans must own. That requires knowing what you’ve deployed, understanding what it’s doing, and making deliberate choices about where the human stays in the loop.

---

I’ve been writing about AI and ROI for a while now, and I keep returning to the same conclusion: we’re obsessed with the wrong variable.

The discourse focuses relentlessly on capability. How smart are the models getting? Which benchmarks are they beating? What can the new release do that last month’s version couldn’t? These are interesting questions for researchers. They’re the wrong questions for leaders.

The question that actually matters is deployment. Where should AI go? What should it be allowed to decide? How will we know if it’s creating value or just creating activity? These are judgment calls that require human wisdom, organizational clarity, and honest assessment of what we’re trying to accomplish.

The companies that figure this out will dramatically outperform the ones chasing model releases. The individuals who develop judgment about human-AI collaboration will be far more valuable than those who simply know how to write better prompts.

One billion agents are coming. They’ll transform how we work in ways we can’t fully predict, just as steam and electricity and computers transformed the generations before us. The pattern of history suggests that the jobs will change rather than disappear, that new work will emerge to replace what automation makes obsolete.

But that transformation won’t manage itself. Someone needs to be driving.

The question for every organization right now is simple: do you know what’s already running inside your walls? And if you don’t, who’s making the decisions about what happens next?