Humans First in the AI Era

Zack Exley’s four-part series, Why Capitalism Can’t Survive AI, is the rare piece of economic writing that seems to want to be wrong. Across four essays it sketches a future in which artificial intelligence does not merely automate individual tasks but staffs the new industries that were supposed to absorb the workers it displaced — breaking a pattern that has held in every prior wave of automation since the cotton gin.

The whole series is worth reading on its own terms — part 1, part 2, part 3, part 4. But here, in one paragraph, is the shape of the argument:

Every previous wave of automation displaced labor and then created new industries that re-employed labor. AI is the first wave that can also staff its own new industries. Market competition forces firms to adopt it whether they want to or not. High-end knowledge work goes first — law, accounting, software, marketing — and when those earners stop earning, a self-reinforcing demand spiral takes the rest of the economy with it. Government stimulus has historically worked as a bridge over a temporary gap; under permanent displacement, the bridge has no other side. Exley argues the arithmetic eventually forces a structural answer — public ownership of productive capacity — because every other path runs out of either money or trust.

Whether Exley is right about the timeline matters less than what his argument exposes about the moment we are already inside. Something is being decided right now, in product roadmaps and procurement decisions and quiet headcount conversations, about what humans are for in a world where machines can perform most of the things humans used to be paid to do. That decision is happening upstream of any UBI debate. It is happening in the rooms where software is shipped.

What this era is doing to people

Strip away the macroeconomics for a moment and look at what the AI era feels like at human scale.

A lawyer who has spent fifteen years learning to write briefs watches a model write a passable brief from a paragraph of prompt. A marketing department of eight is told the new tooling means a team of two can hold the same surface. A junior engineer’s first job is no longer to ramp on a codebase — it is to review what a copilot produced and try to catch the mistakes a senior would catch. The same model that promises leverage to the people at the top of an organization promises substitution to the people one rung lower.

What is at stake here is not only employment. It is the social contract that has, for several generations, made work the primary source of identity, dignity, community, and the daily sense that one’s labor contributes to something. Strip work of that weight too quickly and there is no ready replacement. Hobby and pastime are not synonyms for purpose. A society that learns to live without paid labor as its organizing principle has to learn a new organizing principle — and the work of inventing one cannot be outsourced to the same systems that made the old one obsolete.

This is the part of the AI conversation that earnings calls do not address. It is also the part that will determine whether this era is remembered as a great unburdening or a great abandonment.

The conversation happening upstream

There is a version of the AI conversation that is really a conversation about cost. How many engineers can one prompt replace? What is the headcount delta? How do we get the SaaS multiplier without the seats? That conversation is happening in earnings calls and on enterprise sales floors right now. It will not stop, because, as Exley correctly observes, market competition will not let it.

But there is a second conversation, quieter, happening at the same time in the rooms where software is actually built. Who is the human in this loop, and what does the loop owe them?

That second conversation is the one that decides what the AI era feels like from the inside. It is not a question of policy or arithmetic. It is a question of design — of how systems are configured, what they are permitted to do, what they are required to explain, and whose accountability they preserve. A design ethic does not solve the structural problem Exley names. But the structural problem will be lived out, day by day, inside the products being built right now.

Four principles for the era

If “humans first” is going to be anything more than a slogan, it has to translate into design decisions. Four principles, written down so they can be held to.

1. AI as leverage for people, not substitute for them

The most important question a feature can be asked, before it ships, is: does this make a human more powerful, or does it make a human optional? The same model, the same tokens billed, the same architecture diagram — completely different things on the human side.

A tool that lets a marketer turn one campaign into five, on her terms, is leverage. A pipeline configured to let a company fire the marketer is substitution. Both will sell. Only one of them counts as building.

2. No “scalable human equivalent” thinking inside teams

Exley introduces a phrase — Scalable Human Equivalent — that captures something specific and dangerous: an AI system configured to do the work of an entire team simultaneously. The moment an organization starts sizing its team around what an SHE could replace, it has stopped leading people and started doing inventory on them.

Decisions about AI tooling and decisions about who is on the team should not appear on the same line of the spreadsheet. When they do, the team has already been broken, whether anyone has been laid off yet or not.

3. Software for human readers, not just for machine readers

Designing for AI consumption — search engines, language models, autonomous agents — is real work and not going away. But the second a product is optimized only for the machine reader, the human reader has been demoted to a second-class user on the surface they used to own. That trade-off shows up in writing voice, in design density, in information architecture.

It is also bad on its own terms: the models are getting better at recognizing copy written for them rather than for people, and devaluing it. Write for humans first. Verify the machines can follow. The order is the ethic.

4. The seam between human and AI is the design surface

The most consequential design work in agentic systems is not at the model and not at the UI. It is at the seam between them — where decisions get made about what the model is allowed to see, what it is allowed to do, who reviews its work, and what gets logged.

That seam is where humans either remain accountable for what a system does, or quietly disappear from the loop while still being blamed for the outcomes. The seam should be treated like a contract, not a vibe. Policy is data. Non-engineers can read it. Every decision the model makes leaves a trail. The grammar gives the seam away: a team in trouble says “the AI decided”; a team holding the seam says “we decided, and used AI to execute.”

What is being decided right now

Exley’s argument is structural, and structural problems are not fixed by individual virtue. If the arithmetic runs the way he draws it, the answer eventually has to be a different economic substrate underneath all of us — one not built on the assumption that wages are the only legitimate way to share what an economy produces.

But that larger conversation, about ownership and arithmetic, is downstream of a smaller one that is already happening. Inside every product shipped this year, a choice is being made about whether the AI era gets organized around what people lose or around what people are now finally free to do. The first framing produces a future of permanent triage. The second produces something humanity has never quite had a chance to try.

Humans first is not a brand and it is not a strategy. It is the working version of that choice — made every day, in product reviews and procurement meetings and one-on-ones, by people who often do not realize a civilizational decision is moving across their desk.

Exley’s series argues the larger conversation is coming whether we want it or not. The smaller conversation is happening right now. One will need patience. The other needs answers today.