The Human API

In 1937, Ronald Coase asked a simple question in The Nature of the Firm: if markets are efficient, why do companies exist?

His answer was transaction costs. Finding prices, negotiating, and coordinating are expensive. Firms exist because handling that inside a hierarchy is cheaper than on the open market.

Enterprise software runs on that kind of friction. Turning a customer's problem into working software has always been slow, so a whole class of jobs grew up to manage the handoff.

I call them the Delegators. Strategists, coordinators, program managers, alignment-drivers. They don't write the code and rarely carry a quota. Their value is acting as a human API between the people with problems and the people who build.

The Copilot Trap

Most enterprises still treat AI as a copilot. Same org chart, everyone a bit faster. The seller hands requirements to a translator who writes tickets faster, who hands them to an engineer who writes code faster.

Delegators look great in this picture. Velocity, story points, tickets closed: all up. But a faster assembly line is still an assembly line.

The real shift is agents that do the work end to end. Take a customer who needs their old CRM export connected to your API. The old path is weeks of decks and ROI debates while engineering time is rationed. The new path: someone who understands the customer describes the integration in plain English, and an agent builds the pipeline, writes the tests, and deploys the endpoint.

The translation layer is the part that disappears.

More Software, Not Less

The common fear: if agents build everything, won't we run out of work?

The Jevons Paradox says no. When a resource gets cheaper to use, we use more of it. Cheaper code means far more software, much of it built for one job and thrown away.

When production stops being the bottleneck, value moves to two scarce things: intent and governance. Those are where people should move.

The Zero-Distance Operator

Jim Barksdale put it this way: "There are only two ways to make money in business: one is to bundle; the other is unbundle."

Software work was unbundled. A seller found the pain, a strategist shaped the solution, an engineer built it. Agents rebundle those roles into one person who owns intent: the Zero-Distance Operator.

People with both used to be rare. Agents lower the technical floor, so the gate now is curiosity more than years of coding.

Operators don't need buy-in to build a feature. They build it and show it.

When Friction Is a Feature

Then there's the security team. To a CISO, an operator spinning up an unvetted endpoint that ingests customer data is a breach waiting to happen.

In regulated companies, some friction is the job. Those weeks of debate also cover threat modeling, access control, and resource allocation. Code got cheap. Accountability didn't. You can't sue an algorithm, and an agent can't settle a turf war over a legacy department's API.

If everyone ships disposable apps, shadow IT explodes. How do you audit an app that no longer exists? How do you know it didn't log customer data? Who decides what gets compute when every idea can ship?

The Middle Crystallizes

The middle isn't vanishing. It's crystallizing.

Pure translation is a dead end. A new coordinator is emerging: one who owns AI governance, risk, and priorities across teams, and builds the guardrails that let operators move fast without putting the company at risk.

If your job is coordinating other people's work, rather than facing the customer, directing the agents, or managing risk, your moat is drying up.