You’re Not a Developer Anymore. You’re an AI Dispatcher.

I spent 15 minutes staring at a blinking cursor this morning. Not because I was thinking. Because I was waiting for a machine to finish typing.

It’s absurd. You’re a highly paid professional, and your primary bottleneck is now the speed at which a large language model can regurgitate tokens. The compiler used to be the enemy. Now it’s the API response time.

I asked a room full of developers what they do while waiting for their AI coding agents. The answers were a window into a new kind of corporate hell: “I switch to another prompt. I’m running three tickets in parallel. I’m researching the next ticket to reduce context and token cost.” They’re not coding. They’re managing a queue of AI workers, each one slower than a junior dev on a bad day.

This is the dirty secret of the AI revolution: we didn’t eliminate idle time. We just shifted it. Instead of waiting for compilers, we’re waiting for tokens. And the human—the expensive, creative, strategic human—has become the idle resource, the one bouncing between tasks, the one who can’t focus because the machine isn’t done yet.

I call this the Dispatcher’s Dilemma. Every time you submit a prompt, you become a middle manager. You have to keep the AI busy, or your productivity looks bad. So you pile on more prompts, fragment your attention, and end up doing shallow work on three problems instead of deep work on one.

Here’s the golden quote: “We didn’t replace developers with AI. We replaced developers with AI dispatchers and token-cost optimizers.” And the cost isn’t just API fees. It’s cognitive load. It’s the death of flow. It’s the quiet anxiety of being a human waiting for a machine to finish its homework.

I saw this firsthand at a startup where the CTO bragged about “10x productivity” from agents. But when I walked the floor, every developer had three terminal windows open, each running a different agent, each waiting. They were playing a game of musical chairs with their attention. One engineer told me: “I’m not a developer anymore. I’m a prompt engineer who also reviews code. But I don’t have time to review—I’m too busy queuing the next prompt.”

Take a side. I’ll take one: this is dangerous. Not because AI doesn’t work—it does. But because we’re optimizing for token throughput instead of human cognition. The real bottleneck isn’t the model. It’s our ability to hold a single thought longer than 90 seconds. And every time you switch to another agent task, you pay a switching cost that no model can recover.

The twist? The best developers I know are doing the opposite. They’re not running three agents. They’re running one, and they’re using the downtime to think. To read the code. To draw on a whiteboard. To let their subconscious chew on the problem while the AI grinds through the syntax. They treat the waiting time as strategic, not wasteful.

But most of us don’t. We fill the gap with more work. And that’s the trap: AI promised to free us from drudgery, but instead it’s turned us into frantic dispatchers, optimizing for the wrong metric—machine utilization—at the expense of human depth.

So the next time you’re waiting for your agent, stop. Don’t open another ticket. Don’t research the next prompt. Just sit there. Let the machine wait for you for a change. Because the most valuable thing you can do while the AI generates code is remember that you’re not its dispatcher. You’re the one who decides what to build.