We build with AI every day. We teach with it, ship with it, and write courses about it. So when we lay out where we stand, understand that this isn’t the nervousness of people watching from the bleachers. It’s the position of people with their hands in the tools, people who like a great deal of what they find there, and who think that’s exactly why the rest of this needs saying.

AI is here to stay. That isn’t a hope or a fear. It’s an observation. The productivity is real. In our own work, models help juniors read unfamiliar code, help seniors draft and review, and compress the boring middle of a hard problem. A teaching company that pretended otherwise would be committing malpractice. We are not here to wish it away.

But “here to stay” is not a moral verdict. A thing can be permanent and still be built badly. Our position is simple to state and harder to live. We are for AI that serves human flourishing, and we are skeptical of AI built for no reason but to make money. Here is what that means in practice.

The test is human flourishing

We have one criterion, and we apply it the way we judge code, by what it does downstream rather than by how clever it looks. Technology that makes people more capable, more connected, and more employable is good. Technology that hollows people out, concentrates power, and pushes its costs onto others is not good, even when it is wildly profitable. Every belief that follows is that one test, applied to a specific question.

Cover your costs

AI companies should cover the true cost of what they consume, including power, water, and a fair share of the public purse, in the communities they choose to operate in.

The bill is not small. The IEA reports that data centers used about 460 terawatt-hours of electricity in 2022, roughly two percent of global demand, and projects that number could pass 1,000 terawatt-hours by 2026, about what the entire country of Japan consumes. Too often the public underwrites the buildout. The Union of Concerned Scientists found that ratepayers across seven states in the PJM grid were on the hook for 4.3 billion dollars of infrastructure approved in a single year just to connect data centers. Georgia estimates it forgoes around 2.5 billion dollars a year in data center tax breaks, with Virginia and Texas not far behind. By 2026 the backlash was loud enough that Illinois, Arizona, Ohio, and others began pausing the incentives.

A company that books the profit and offloads the electric bill, the water table, and the tax base onto a town of strangers is not innovating. It is externalizing. We think good environmental stewardship, honest utility arrangements, and a fair tax base are the price of admission, not a line in a press release.

Power and water, as things stand, are not yet sustainable

We will say the uncomfortable part plainly. The current trajectory of resource use is not something the grid, the water supply, or the climate can absorb indefinitely. Roughly two-thirds of the data centers built in the United States since 2022 have gone up in water-stressed regions, according to a Bloomberg analysis, including the deserts of Arizona. We don’t raise this to scold. We raise it because pretending otherwise is how you earn a reckoning instead of a transition. The companies that solve efficiency and clean power will deserve the lead they get. The ones betting that someone else will solve it later are borrowing against a future they do not own.

Pay the people whose work trained the models

Models learned to write by reading what humans made: books, code, articles, and art. The people who made that work deserve fair compensation for it.

This is no longer a fringe view. It is moving through the courts in real time. The New York Times is suing OpenAI and Microsoft, and in 2025 a federal judge let the core copyright claims proceed. Anthropic agreed to pay about 1.5 billion dollars to settle a class action from authors whose books were used without permission, roughly 3,000 dollars per title across an estimated half million books. We don’t need the final ruling to know the principle. If a person’s work made your model more valuable, they are owed something. Build that in early. Don’t litigate it out later.

Vibing is counterproductive. Judgment is the job.

This is our most practical belief, and the one we stake our teaching on. Pointing an agent at a problem and shipping whatever comes back, which people now call vibing, feels fast and fails slowly. On anything that has to live longer than a demo, judgment is the scarce and decisive skill: knowing what to keep, what to throw away, and where the design is quietly wrong.

AI raises the ceiling for people who have that judgment and lowers the floor for people who do not. The gap between those two groups is the whole game. We spend our time teaching the judgment, because it is the part the tools cannot hand you.

There is still a place for juniors, and the dividend should pay to train them

The cynical reading of the AI dividend is that we can stop hiring juniors. We read it the other way around. Some of what AI saves should be reinvested in the people who become your seniors in 2030.

Cut the training pipeline and you eat your seed corn. You get a few strong quarters and then a hollow team with no one coming up behind it. The junior role did not vanish. It moved up the stack, toward planning and review, and it needs more mentoring now, not less. A healthy organization spends part of its dividend on the next generation. That is not charity. It is the only way the senior bench ever refills.

Human connection, creation, and work matter more now, not less

As machines do more of the producing, the human parts go up in value, not down: judgment, taste, relationships, and the simple act of making something and standing behind it. We are a small teaching company. We believe people learn from people. AI is a powerful tool in that room. It is not a replacement for the room.

In short

We believe in AI that serves humanity. AI that pays its own way, compensates the people it learned from, sharpens human judgment instead of substituting for it, and leaves room for the next generation to grow into the work. We do not believe in AI built solely to make money, with the costs shipped off to someone else’s town, grid, or career.

That is where we stand. We intend to build, teach, and hire like we mean it.

Sources

• Data center electricity use (460 TWh in 2022, projected to top 1,000 TWh by 2026): IEA, Electricity 2024 · DatacenterDynamics summary
• Ratepayers and tax breaks ($4.3B PJM grid costs; Georgia/Virginia/Texas exemptions; 2026 rollbacks): Stateline · Good Jobs First
• Data centers in water-stressed regions (Arizona; Bloomberg analysis): Grist · Tom’s Hardware
• Anthropic $1.5B author settlement (Bartz v. Anthropic): NPR
• NYT v. OpenAI/Microsoft, copyright claims allowed to proceed (2025): Axios

— Bruce