June 2026
A few weeks ago an investor said something on a call that I haven’t been able to put down: “Humans are the rate limit on AI.”
He’s right, and almost everyone has the story backwards. The default narrative is that AI shrinks the need for people. Then you try to deliver real outcomes with agents, not demos, and you arrive at the opposite conclusion.
I’ve watched team after team try to automate a process end to end, sales, support, research, and hit the same wall. The agent does 90% of it beautifully, then falls over on the part that actually matters: the live sales call, the judgment call, the moment a customer needs a real person. The agentic companies that are winning aren’t the ones that removed the human. They’re the ones that put a human and an agent together and sold the outcome.
But here’s the catch. A full-time employee is far too expensive and far too rigid to sit in that loop. So if humans are the rate limit, then how we hire, allocate, and pay them has to change completely. That’s the thesis behind everything we’re building at 8x.
There’s a 160-year-old idea that explains this better than any AI take written this year. In 1865 the economist William Jevons noticed that when steam engines got more efficient, England didn’t burn less coal. It burned far more. Make a resource cheaper to use and you don’t use less of it; you find ten new reasons to use more. Economists still call it Jevons’ paradox.
It’s exactly what AI is doing to work.
When an agent can do 90% of a workflow for almost nothing, the job doesn’t disappear. The cheap 90% multiplies the surface area of the expensive 10% that still needs a human. Look at sales. AI now writes the emails, books the meetings, drafts the proposals, updates the CRM. So a team can run ten times the pipeline it used to, which means you need more closers on the phone, not fewer. a16z put numbers on this. In From ‘System of Record’ to ‘System of Intelligence’ they asked sales leaders how AI will change their headcount over the next two years: 39% expect to grow their teams, 37% expect it to stay flat, and only 5% expect a serious cut. In their words, “the ROIs on these agents are strong enough that the total pie grows rather than the labor budget shrinking.”
This is the part the headlines miss. Agentic loops finish faster than any one person can supervise. Claude Code closes a task while you’re on another tab. Compute stopped being the constraint a while ago. Human attention is the constraint now. The companies that win the next decade won’t be the ones with the best model. They’ll be the ones who can put the most humans around the most loops, fast.
Here’s the version of this that still surprises me, and I run the company. Watch what happens when a brand switches from human-made content to AI-generated content. You’d assume the people disappear. They don’t. The headcount roughly doubles.
Generated video doesn’t arrive finished. Someone has to edit it, fix it, caption it, and localise it so it lands in each market, and that production work goes to people in cheaper labour markets. Then someone still has to post it. You can’t bot your way past TikTok and Instagram; the algorithms are built to bury automated, bulk, low-trust posting and to reward real accounts with real devices and real behaviour. So you need a person in the US holding an actual phone with a warm account just to get the AI content distributed at all.
One creator becomes two roles: a producer who makes the content and a poster who carries the trust. The most automatable workflow in our entire business turned out to be the one where AI added people instead of removing them. If the paradox holds there, it holds everywhere.
If the whole game is putting humans around agentic loops, the first question is brutally practical: how do you get millions of humans onto a platform in the first place?
There’s only one channel with that kind of reach, and it’s social media. And the timing has never been better. The job market is fundamentally broken. Smart, hungry people everywhere are looking for a way in and can’t find one through the traditional door. That’s the window: you can acquire humans at scale, put them on an app, upskill them, evaluate them, and then allocate them across agentic workflows where they actually deliver value.
That’s why we started with creator networks. It’s the perfect wedge. Companies already know that orchestrating hundreds of creators is hard, so they’ll pay for it on day one, and it performs in the one place performance is undeniable: marketing. We get immediate market pull. But the same machine that recruits, screens, briefs, and scores a creator is quietly laying the rails for human orchestration at scale.
We have 121,000 people on the platform today, and we acquired them through our own viral content, not ads. We’re our own first case study: the same loop that gets a brand 24 million views gets us our next ten thousand workers.
Once you have an engine that can acquire, evaluate, and rank humans at scale, the question stops being “which marketing problem do we solve” and becomes “which kind of human work do we plug in next.”
The infrastructure that finds the top 5% of creators is the same infrastructure that finds the top 5% of salespeople, or engineers, or annotators. It runs on a power law: in every vertical we’ve touched, a small fraction of people drive most of the value. And for the first time in history you can actually act on that at scale. You can upskill ten thousand people at once, evaluate ten thousand people at once, run ten thousand people through the same live test at once, and keep the few who break out. AI is what makes that possible. No agency, no staffing firm, no platform could do it before.
We already have 8,400 engineers on the platform and live work in sales and data, all run by a core team of six or seven people. Everything past that core is outcome-based and on demand: you don’t hire the team, you buy the result.
We’re not the only ones who’ve worked out that humans are the bottleneck. Mercor just raised at a $10B valuation and pays its expert network more than $2M a day. Surge AI crossed $1B in revenue. The smartest capital in AI is pouring into human-in-the-loop work, and the same power law runs through their business too: roughly 10% of the annotators produce 90% of the value.
But their real constraint isn’t finding the good ones. It’s acquiring humans at all. They’re recruiting PhDs and doctors and engineers one expensive hire at a time, and they don’t own a channel that brings people in on its own. We do. Our creator networks are an acquisition engine we run on ourselves, and they bring people onto the platform for cents, disproportionately cheaper than buying them through Facebook ads. When the bottleneck of the entire industry is human supply, owning the cheapest pipe to that supply is the moat.
There’s a second thing this engine produces, almost as a byproduct: data.
Every workflow we run, every human in every loop, generates a record of how the hard 10% actually gets done. That matters more than it sounds. The only way a model gets better at a task is data from the gaps it can’t yet cross, and we’re sitting on exactly that, on the most valuable last-mile work, across verticals. When the next model lands and a fresh slice of that work becomes automatable, whoever holds the most data on those use cases is the one who takes them over.
That’s the shape of the next decade. The last one was about compute: who had the models, the training, the chips, and how much value you could squeeze out of an agent. The next one is about who can build the systems that orchestrate humans and agents together to deliver real outcomes in the real world, and who collects the data those systems throw off.
People will keep building harnesses around agents to take the human out of the loop, and they’ll succeed, one gap at a time. But here’s the counterintuitive part, and the reason I think this model is future-proof: every time a gap closes, a new one opens. There is always a fresh last 10% that only a human can do, and someone has to acquire, train, and allocate the people who do it.
Our answer is a billion of them. That’s the company we’re building.