We'll All Run Faster and End Up in the Same Place
What happens when everyone has AI agents racing on their behalf? A few possible futures, who pays for them, and the business hiding in each.
Jacob RogelbergWe're already there. People are handing goals to AI agents and letting them run. Book the table. Reschedule the call. Chase down the refund. Big consumer platforms are shipping this to normal people right now. Not as a demo. As a feature.
And it's going to speed up. Soon everyone has an agent. Some people will have thousands. All of them doing things at lightning speed, at the same time, in the same places.
I keep wondering what that world actually feels like.
The cost we never noticed
Most of our systems quietly assume that bothering someone takes effort. A reservation line. A support queue. A checkout page. They work because people won't spam them. Spamming is annoying. It takes time.
Agents delete that cost. Pinging becomes free and infinite.
You can already see the cracks. Give an agent a goal like "book me a table" and it might hammer an API, or find some creative workaround nobody intended, because the goal is the goal. On the other side, platforms start banning accounts because something is calling way too often. Was that you? Your agent? The company that built it? We don't have good instincts yet for delegated intent at machine speed.
Faster, and nowhere new
Here's the part that bugs me most.
Every individual decision makes sense. My agent gets me the table faster. Your agent gets you the appointment first. Short term, everyone wins.
But when everyone has one, the advantage evaporates. Agents become the baseline cost of participating, like email. The racing cancels out. We all end up in the same place, just more exhausted, with more noise in every system we touch.
We all end up in the same place, just more exhausted.
It's a tragedy of the commons. Except the commons is attention, and everyone's agent is grazing on it.
Running it forward
So I tried to think about it like a Monte Carlo simulation. Spin the next six to twelve months a bunch of times. Where do we land?
The boring equilibrium. Platforms build official lanes for agents. APIs, rate limits, some kind of verified-agent badge. The chaos gets zoned off, the way spam mostly got tamed by filters. Duller than the hype. Probably the most likely outcome.
The arms race. Detection versus evasion escalates. CAPTCHAs creep into everything. The irony: humans suffer most, because we're the ones who fail the "prove you're not a bot" tests. The agents pass fine.
The trust collapse. A few ugly incidents. An agent drains an account. A swarm of them knocks over a booking system. Platforms panic and ban automation outright. Regulation shows up late and clumsy.
It actually works. Agents negotiate with agents. Protocols emerge. Machine-to-machine commerce just hums along and we barely notice. Least likely in the short term. Quietly the most interesting.
Who can afford it
None of these futures hit everyone the same way. It depends a lot on how big you are.
The giants. The big platforms set the rules. They can afford detection systems, verified-agent rails, smart rate limits. The agent flood is a problem they can engineer around. It might even help them. Complexity is a moat, and smaller competitors can't match the infrastructure.
The middle. Mid-size businesses won't build any of this. They'll subscribe to it. Whatever agent-handling layer the market produces, they'll plug it in and move on. Fine, as long as that layer exists and stays cheap.
Your hairdresser. The corner café. The solo plumber. One shared calendar, a phone, maybe a booking link. When a hundred agents start pinging for the last open slot, they have no defense and no budget for one. They either shut off automation entirely, or they get gamed.
The people with the least capacity to adapt meet the chaos most directly. Agents don't level the field. They tilt it toward whoever can afford the countermeasures.
Where the openings are
Every one of these futures needs something that doesn't really exist yet.
| If we get... | Someone needs to build... |
|---|---|
| The boring equilibrium | Rails for agents. Verified identities, permissioned actions, rate-limit middleware. |
| The arms race | Trust layers. Proof that a human is human, and that an agent is authorized to act. |
| The trust collapse | Liability tooling. Audit trails, and some form of insurance for when an agent goes rogue. |
| It actually works | Protocols and marketplaces where agents negotiate with each other. |
The big players will build most of this for themselves. The more interesting question is who builds it for the hairdresser.
I don't know which future we get. Probably some messy blend of all four.
What I do know is we're building it right now, mostly without asking where it goes. Everyone's optimizing for the next booking, the next reply, the next small win.
So where does all this go? And if you were building for it, which future would you bet on?