Two people ran the workload of six teams. The claim sounds like a stretch, which is exactly why it needs setting up before the receipts arrive. At Machintel, two people carried the surface area of six teams, the work landing roughly ninety days after the system went live, against two years of earlier experiments that never stuck. This is the short version of that story, told to make the shape of it believable first. The full account, with the numbers and the parts that did not work, is the standalone case study this points to.

Not by working harder. By encoding the judgment, the plays, and the handoffs into the system so that the operation and the team carried the surface area, rather than two pairs of hands working longer hours. The work that used to need six teams did not disappear and it did not get done by two heroic people. It got carried by the way the work was set up to run.

That is the distinction the whole story turns on. A business at this stage usually scales surface area by adding headcount, because the know-how lives in people’s heads and the only way to cover more ground is more heads. Machintel covered the ground a different way: the know-how moved into how the work ran, so the team could carry far more of it without the count going up. Revenue per employee ran two to three times what it had been, which is the measurable shadow of that shift.

Because the wider market is already running well past it. Tiny teams are posting output that used to require headcount nobody at that size could afford, and investors have started treating revenue per employee as the efficiency signal that matters most. One account of the trend documented a twenty-person company at five million in ARR and a sixty-person platform at roughly three million dollars in revenue per employee, with large headcounts at AI-native companies now read as a warning sign rather than a strength.

Set the Machintel range against that band and it lands inside it, not beyond it. The owned two-to-three-times figure is deliberately conservative next to a market where the leading edge runs several times higher. That is the point of holding the story up against the trend: the claim is not an outlier, it is the middle of a documented shift, which is what makes it believable before you have seen a single internal number.

The judgment and the handoffs. The recurring decisions that used to require a person who knew the business got encoded as plays the system could run, and the points where work passed from one stage to the next, the places where things usually fall through the cracks, got built into the operation instead of living in someone’s memory. That is what let two people cover the ground of six teams: most of the surface area was carried by templated judgment, not by manual effort.

The full breakdown of which decisions got templated, and how, is the substance of the complete story. The short version is that the leverage came from encoding the know-how, not from speeding up the typing.

The honest answer has two halves, and the second one matters as much as the first. The recurring, rules-bound work, the qualification, the follow-through, the handoffs, stopped being done by hand and got carried by the system. The judgment calls that genuinely needed a person stayed human, on purpose.

That second half is what separates this from the version of the story nobody should believe. There is a well-known pattern of companies pushing AI to do the work without the team running it and then quietly walking it back. The Machintel story avoided that failure mode precisely because it kept the team in the operation. Two people running six teams’ worth of surface area is a story about a team carrying more because the system carried the rest, not a story about AI replacing people.

This is a teaser, so it sets up rather than proves. The reason to read the full story when it lands is that it is the honest version: it names the two years of experiments that did not stick before the ninety days that did, and it says plainly what did not work alongside what did. Machintel ran on Works, the operations layer that held the context, the plays, and the handoffs, which is how two people and a team carried what used to take six. The receipts live in the complete account.

The full Machintel story is coming. In the meantime, sign up for early access to see the operation it ran on. The ROI-angle companion, the receipts side of the same story, is in Machintel receipts, and the two years of experiments that came before are in why I spent on AI and failed anyway.

Two people carried the surface area of six teams. Ninety days on the system, after two years of experiments. Believe the shape of it first. The full story is what proves it.

How can two people run the workload of six teams without just working harder?

Two people covered six teams’ worth of surface area because most of that surface area was carried by the system, not by their hands. Know-how that previously lived in people’s heads moved into how the work ran, which is the only way to cover more ground without the headcount going up. Revenue per employee landing at two to three times the prior rate is the measurable shadow of that shift: the gain was in the encoding, not in the hours.

What did we actually template, and why did it matter?

Templating shifted the bottleneck from “a person who knows the business” to “a system the team can run.” The two categories that moved first were recurring decisions with a knowable right answer most of the time, and handoff points where work typically fell through the cracks between stages. Both had been carried in someone’s head. Moving them into the operation meant two people did not have to re-carry the context on every cycle.

What is the honest version of this story, including what did not work?

Two years of earlier experiments produced tools the founder operated alone, not operations the team could run. None of them compounded. The ninety days that worked came after that failure because they started with the team carrying the system rather than one person owning it. Encoding judgment and handoffs so the team could hold more surface area was the move the earlier experiments skipped. The complete account, with the receipts, is the standalone case study this piece points to.

Why is the two-to-three-times revenue per employee figure conservative?

Because the documented market band already runs well past it. Tiny teams at the current leading edge of AI-native companies post revenue per employee that makes the Machintel range read as the middle of a shift, not an outlier. Holding the owned number to the honest range, rather than inflating it to match the highest-profile examples, is exactly why the claim is worth making: it lands inside what the evidence already shows is happening.

1. Sifted, “Tiny teams, big revenue, but for how long?,” 2025
   
   

Keep reading:

• Machintel receipts
• Why I spent on AI and failed anyway
• Team Runs With Me