You Are the Human in the Loop
The panel agrees the human in the loop becomes the premium. They are right. They skip the only question that matters: which human.
On the July 3rd All-In, the four of them landed on something they rarely land on together. Automation is coming for the bottom of the org chart, the entry-level and the repetitive, and the counter-move is not to fight it. It is to become the human in the loop, because that human turns into the premium.
They are right about the trend. They are pointed at the wrong human. On the podcast, the human in the loop is the concierge, the bartender, the driver, the customer-service rep you can still reach. That human matters. But it is the small version of the idea. The human in the loop who decides whether your company is still standing in three years is not at the front desk. It is in the chair where the work gets directed. That is the one this brief is about.
What they got right
The panel spent a stretch of the episode circling the same point from different directions, and for once nobody dug in against it.
"The counternarrative to automation generally is that we're going to realize the importance of human interaction and humans in the loop, and we're going to pay a premium for it."
"Human in the loop will be the premium. We will all realize in the next year and a half the importance of human in the loop, even in the context of AI and automation working."
The All-In panel · Jul 3, 2026 · ~44:00They reached for the obvious evidence. The company that announced it would replace its whole customer-service department with AI, made a lot of noise about it, then quietly walked it back a year later because customers need to know a human is there if they want one. The construction crew of the near future: twenty humans, fifty robots, and the humans telling the robots what to do. Their word for those humans was supervisors. And the working data from Ramp: the companies that actually use AI are not shedding people, they are growing faster and hiring more.
Every one of those examples is true. Every one of them is also about somebody keeping a human in the loop. Nobody on that call was describing a business with no humans in it. They were describing where the human moves to.
Not the IT department
Here is where most companies get it exactly backwards. They decide AI is a technology, so they hand it to the technology people. It becomes a project. It gets a budget line, a pilot, a committee, and a quarterly update. And it goes nowhere, because the person who has to sit in the loop is not the person who administers the software. It is the person who already knows what good looks like.
Where it goes to die
- Treats AI as a tool to deploy, not judgment to direct
- Knows the systems, not what a good outcome is
- Runs a pilot, files a report, moves on
- Cannot make the call the machine can't
Where it belongs
- Already carries the standard for what good looks like
- Has the context, the scars, and the authority to decide
- Reads the output and knows instantly what is wrong
- Owns the outcome, so owns the loop
AI does not know what your business is for. It will not figure out what matters. That judgment is the one thing that has never lived in IT, and it is exactly the thing the loop runs on. Which means the loop cannot be delegated to the org's most junior technical function. It sits with the people who have been making the hard calls all along.
The kid straight out of grad school
If you want to know what you are actually working with, here is the most useful picture I have found.
AI is the smartest kid you ever hired straight out of grad school. Smarter than you at every discrete thing on the desk. Writes faster. Reads faster. Recalls everything. Never tired, never bored, works through the night without a word of complaint. On raw horsepower, this kid outclasses you and it is not close.
And also: no judgment. No context. No scars. No idea what your business is for or where the landmines are. Give it a bad instruction and it will execute the bad instruction at full speed and hand it back to you with total confidence. Left alone, it will do the wrong thing beautifully.
You already know how to manage this person. You have done it for thirty years. You do not learn a new tool for it. You guide. You usher. You correct. You keep them pointed at the goal they cannot see yet. The muscle is not technical. It is the oldest management muscle there is.
Now multiply the kid
One brilliant grad is a novelty. Here is the part that changes the shape of the job: you do not get one. You get ten. Or a hundred. Whatever your tolerance is for open terminal sessions running on your own computer.
As I write this, I have twenty-four of them open. Each one is a session doing real work, drafting, reconciling, researching, building, all at once. My leverage is no longer how much I can personally produce in a day. It is how many capable things I can keep pointed in the right direction at the same time.
Solid: work comes back to be read and corrected. Dashed: direction goes out. One operator, many sessions. Your whole job is the ring.
The number is not the point. Twenty-four is just what fits my head today. Yours might be three, or eight, or forty. The point is that the ceiling on your output stopped being your own two hands and became your capacity to direct. That is a different job than the one most executives think they have.
Read. Correct. Decide. Keep it going.
Strip away the tooling and the workflow is four verbs. This is the whole thing.
You are not doing the work. You are reading what came back, catching what is wrong, making the calls it cannot make, and starting the next one. Run that across a stack of sessions and you are operating at a scale no single team can match, on judgment that no model has.
The work is the machine's. The loop is yours.
This is management, returned
None of this should feel foreign. It should feel like coming home.
The skill the loop demands is the oldest one in the building: take younger, faster, more capable people and guide them toward an outcome they cannot yet see. Read their work. Correct it. Point them again. The best operators always did this, and the ones who were great at it built careers on it before anyone said the word AI out loud.
What changed is only the count and the clock. You are no longer guiding one bright associate over a year. You are guiding a dozen at once, at machine speed, and the correction that used to take a review cycle now takes a sentence. If you were a good manager of people, you already own the muscle this requires. The job did not get more technical. It got more like the job you were already good at, multiplied.
The shift
| Doing the work | → | Directing the work |
| One thing at a time | → | A stack of sessions at once |
| AI as an IT project | → | AI as a management job |
| Output capped by your hands | → | Output capped by your judgment |
| Managing people | → | Guiding capability, human or not, the same way |
The human in the loop is not going away. The panel is right about that. But being in the loop is a skill, and the skill is learning what you are working with.
The sooner you understand that this is your job now, not IT's, not a vendor's, yours, the sooner you are on the rocket ship.
The human in the loop is not going away. But you have to learn what you are working with.