McKinsey's CEO recently said he now counts 20,000 AI agents as employees. Alongside 40,000 humans. Target: 1:1 by end of 2026.
My first reaction wasn't "that's impressive." It was "how do 60,000 workers know who's doing what?"
That question is the whole game right now. Not capability. Coordination.
We Solved This Once Before
In 1920, 8 million cars were on American roads. No stop signs. No traffic lights. No lane lines. No licenses. No speed limits. 15,000 people died that year. The death rate: 18 per 100 million vehicle miles.
Fast forward to 2025. Nearly 300 million vehicles on the road. Cars are faster, highways are denser, people drive farther. The death rate: 1.10 per 100 million miles. A 94% drop.
Cars didn't get slower. We got 37 times more of them. And the fatality rate still fell 94%.
What happened in between was coordination infrastructure. Lanes, signals, right-of-way rules, licensing, speed limits. Connecticut passed the first speed limit law in 1901, just 12 mph in the city. Not because speed was the enemy. Because nobody had built the system for what happens when everyone is fast. The roads didn't just move people. They became the rules that kept people from destroying each other.
AI Is the New Horsepower
Each person on your team just got dramatically more powerful. Individual task throughput is up significantly for many roles. Developers report up to 88% gains with AI coding tools. McKinsey claims it saved 1.5 million hours in one year.
One person can now do what used to take a team and like every previous technology leap, it creates new problems.
For centuries, human work has been sequential. I finish my part, hand it to you, you hand it to the next person. That pace was slow, but the slowness did something useful. It gave everyone time to check direction. To notice when someone was drifting. To course-correct before things went sideways.
AI breaks that. When each person's output accelerates, the work outruns the organization's ability to stay aligned. Someone builds the wrong thing fast. Two people solve the same problem differently without knowing the other exists. The compounding happens in days, not quarters and many large (and small) organizations are already feeling this pain.
This Is Not a Meetings Problem
The instinct is to add more check-ins. More standups. More review gates. More humans in the loop.
That's the wrong answer. It just re-creates the bottleneck you're trying to eliminate and zaps the productivity gains.
The coordination has to be built into the system. For AI agents, that means automated guardrails. I wrote recently about Frase.io and how every agent action is classified as categorically blocked or contextually permitted, enforced automatically. The agents can't override their own rules. The point isn't the specific implementation. The point is that coordination between humans and AI is a system design problem, not a supervision problem.
But here's what most people miss: humans need the same thing. Not guardrails exactly, but the equivalent. When every person on the team can produce at 2x or 5x their previous speed, scope clarity becomes the single most important organizational capability.
Who owns what. Where your work ends and mine begins. What the actual objective is, stated precisely enough that two people working independently arrive at compatible output. These aren't soft skills. They're infrastructure. They're the lanes and signals for human coordination.
Small Teams Already Get This for Free
A two-person startup with AI doesn't have a coordination problem. Scope is obvious. Communication is instant. They can move at the full speed AI enables.
This is why smaller teams and solo operators are capturing the AI advantage faster than anyone. The coordination is built in at that scale.
For larger organizations, that advantage has to be engineered. The company with 50 or 200 people needs to build what the two-person team gets naturally: precise ownership, explicit scope, and alignment mechanisms that move at the speed of the work. Not at the speed of the next planning cycle.
Build the Roads
1920 had 8 million cars, no rules, 18 deaths per 100 million miles. In 2025: 300 million vehicles, 1.10 deaths per 100 million miles. More cars. Faster cars. A 94% drop in fatalities.
The answer was never to slow the cars down. It was to build coordination into the roads themselves.
AI has given everyone on your team new horsepower. The question is whether you've built the roads. Not more meetings. Not more approvals. Actual coordination infrastructure: clear scope, agentic rules where necessary, defined ownership, shared direction precise enough that fast-moving people don't collide.
The organizations that figure this out will move at a speed their competitors can't match. The ones that don't will have very well intentioned people building very fast in very different directions.