Building Companies in the AI Era: How to Think About People, Roles, and Compensation

Every time I walk into a company that wants to talk about AI, the first question is about tools. Which platform, which model, which vendor. I let them ask it, because it is the question they came in with.

The second question, once they have calmed down about the tool list, is about people. Who do we keep. Who do we hire. What do we pay. Who does what now.

That second question is the one that actually matters. The tool question has a thousand reasonable answers and most of them are fine. The people question is where companies in this cycle will either compound or quietly fall behind, and the answers are not obvious.

What 100X Actually Means

I keep hearing “AI is going to make our team 100X more productive.” It sounds aspirational and it sounds wrong.

It is wrong because 100X is not a uniform productivity bump applied to everyone. That framing lets leadership keep the existing org chart, add a few subscriptions, and pretend nothing structural has changed.

What 100X actually describes is one person doing the work of one hundred, because that person is directing systems instead of executing tasks. The lever changed. The ratio between human input and organizational output changed with it.

That is not a layoff strategy. It is a structural shift in how organizations compound. The companies that internalize it will look very different in three years than the companies that did not.

The Three Roles

In a company built around this shift, every contributor falls into one of three structural roles.

Architects. These are the people with the judgment to orchestrate AI and review its output. In engineering, the best engineers are no longer the ones writing the most code. They are the ones who can tell a bad architectural decision from a good one after the agent drafts it. The skill is judgment, not execution.

This role compounds. Pattern recognition improves with reps, and reps require exposure to many problems and many drafts. An Architect who has reviewed a thousand AI-generated implementations is structurally more valuable than one who has reviewed ten — and the gap widens with time, it does not close.

Orchestrators. These people own the AI systems. They built the workflow, they run it, they improve it, they know when it is wrong. New operator role, not a renamed manager role.

The cleanest analogy is manufacturing. A CNC machinist writes the program. The operator runs the machine. Both jobs get paid. But the machinist’s contribution compounds — every program they write keeps producing parts long after the session ends. The operator’s contribution is consumed in the hours they work.

The Orchestrator is the machinist of the AI-native company.

👉 Tip: The people most likely to become great Orchestrators are not the most senior or the most technical. They are the ones who are already annoyed by the inefficiency in their current role and have been quietly fixing it with spreadsheets, macros, and scripts. They have been operating like Orchestrators without the title. Find those people. They are already in your building.

Relationship-Focused People. These are the customer-facing people. The ones who show up, listen, build trust, and are the reason a customer stays.

As AI absorbs more of the system layer, this role becomes more valuable, not less. The goal is not to replace them. The goal is to remove every system task from their plate so they spend close to one hundred percent of their time on the relationship. That is the lever.

This is the role most leaders get wrong. They see AI quality matching or exceeding human quality in some narrow task and conclude the human role is going away. They are confusing the system layer with the relationship layer. AI handles the first. Humans hold the second. The customer who trusts the person across the table does not transfer that trust to an agent on the other end of a chat window. Not yet, and probably not soon.

AI Makes the Org More Reliant on People, Not Less

This is the counterintuitive part, but it is true.

The quality of the judgment at each node determines whether AI multiplies output or multiplies garbage. A bad architectural call by an Architect does not produce ten bad lines of code anymore. It produces ten thousand. A weak Orchestrator does not slow down one workflow — they slow every downstream process that depends on it.

You need fewer people. You need dramatically better people. And you need to keep them.

A 100X organization that loses a key Orchestrator does not lose one person’s worth of contribution. It loses the operational context that person carried, the systems they built, and the judgment they applied to the edge cases the agent could not learn on its own. That is not replaceable in a quarter. It is barely replaceable in a year.

The moat is not the AI. The moat is the organization built around it.

That sentence is the whole article in one line. Every model you can buy, your competitor can buy too. The thing they cannot buy is the team you built, the context that team has accumulated, and the systems that team owns.

How to Think About Compensation

Traditional compensation bands assume a roughly normal distribution of output across a role. The top performer in a band is maybe two or three times the median, and the band is sized accordingly.

That assumption is dead the moment AI enters the org. If one person genuinely produces one hundred times the output of the median by directing systems instead of executing tasks, paying them 1.5X the median is structurally misaligned. The most leveraged person on the team is the most underpaid relative to the value they create. They notice. Eventually they leave.

A few principles.

Outcome-based, not role-based. Pay against what the contributor’s work produces, not the title on their badge. An Architect whose architectural call enables fifty downstream agent-generated PRs is doing something fundamentally different from an Architect who shipped ten PRs by hand.

Asymmetric upside for system builders and managers. The people whose work compounds get pay that compounds with them. A Orchestrator who has owned a workflow for three years is sitting on operational intelligence that is structurally hard to replace. The retention horizon should be measured in years, not vesting cycles.

Differentiated by role. Relationship-Focused People are not rewarded on system leverage. They are rewarded on relationship outcomes and trust accumulation. The principle applies across the org. The mechanics differ by seat.

Long retention as the default. The cost of overpaying a true 100X contributor is bounded — whatever delta you pay above market. The cost of losing one is unbounded, because the replacement may not exist, and even if they do, the catch-up curve is years long.

👉 Tip: If you are setting comp in this cycle, do not start from the question what do other companies pay for this role. Start from what would it cost us to lose this person and replace them. That second number is the right anchor, and it will almost always be larger than the first.

Building the Team

The next question, once the framework is in place, is who goes in which seat.

The instinct for Orchestration is rare but identifiable. Who in the existing org has already automated something in their current job without being asked? Who built the spreadsheet that closes the loop everyone else complains about? Who wrote the macro nobody told them to write? Those people have been doing the job before the role existed. Give them the seat.

Architects are identifiable by their pattern recognition speed. The signal is not how fast they write — it is how fast they spot the wrong approach. Watch them in design review. The Architect reads the draft once and says “the third decision is wrong, here is why,” and is right. Speed of judgment, not speed of execution.

Relationship-Focused People are the people everyone wants to talk to. Customers ask for them by name. Teammates route hard conversations through them. They are usually busy listening.

All three roles exist in every org I have walked into. The question is never do we have these people. The question is are they in the right seats. Most of the time, they are not. The Architect is buried in execution work that should be done by an agent. The Orchestrator is in a job description that does not let them touch the systems they would naturally own. The Relationship-Focused Person is doing fifteen percent relationship work and eighty-five percent system tasks.

👉 Tip: Before hiring a single new role for an AI initiative, audit the seats you already have. Most companies discover they have the people. They have not given them the right work.

The Closing Frame

The companies that figure this out in the next three years are going to compound in ways that are structurally hard to close from outside.

Not because they have better models. Everyone has the same models. Not because they have more data. Most data is commodity. They will compound because they have built teams organized around this framework, compensation that retains the people who matter most, and systems that accumulate operational intelligence specific to their business — intelligence that does not exist anywhere else and cannot be acquired.

That is the operator thesis. AI does not turn a company into a 100X company. The people, the roles, the comp structure, and the systems they build do. The AI just changed the leverage available to them.

If you are running a company in this cycle and the people question feels uncomfortable, that is the right reaction. It is the question that matters.

If you want to talk through how to apply this to your own org — the three roles, the comp architecture, the Orchestrator hiring, the build vs. buy on agents — that is what I do as a fractional Chief AI Officer. The page is here: /caio.