The Man Behind Me Was Spending His Salary on Claude

The rising costs of AI.

Here I was, sitting on a plane on the Philadelphia tarmac, which is the airport’s way of asking passengers to reflect on every bad choice that led them there. The plane had just boarded its priority passengers, including me, because I had paid for it. This is not something I usually do. I am not, spiritually, a priority boarding person. But with everyone now carrying luggage as if they are fleeing the fall of civilization, overhead bin space has become a scarce resource. And when a scarce resource appears, capitalism does what capitalism does best: it charges you extra for the privilege of not suffering quite as much.

Behind me sat two men on their way to Estes Park for a wedding. I was not exactly eavesdropping. They were speaking loudly enough that I felt less like a stranger listening in and more like a reluctant member of their advisory board. One of the men was a developer. He told his friend he was being promoted at work because of his use of AI. Specifically, he had built multiple agents in Claude, an AI tool I know about but have not personally tried. These agents had apparently put him on the fast track at his company.

There was, however, a small problem. Because there is always a small problem. His senior leadership had not funded his AI expenses. They had concerns. Some of these concerns were reasonable, including the possibility of becoming dependent on AI tools that may become much more expensive once everyone is locked in. So the developer was paying for Claude himself. According to him, he was spending around 30% of his own salary on it. He was proud of what he had built, but also worried. If he kept building agents, buying tokens, and waiting for the company to catch up, he might end up putting most of his salary into Claude just to stay ahead.

There are worse career strategies, I suppose. Though not many that sound so much like a tech-themed treadmill where the incline keeps increasing, and the treadmill sends you an invoice. By that point, the plane was fully boarded. To my relief, both men eventually decided to take a nap. I was grateful. I like my transportation quiet. I do not need every flight to become a podcast.

But the conversation stayed with me. It made me think about AI, work, money, skill, creativity, and the general direction of things, which, admittedly, is a lot to put on two men trying to get to a wedding in Colorado. Still, I think the conversation revealed something important.

AI is going to widen the gap between people who know what they are doing and people who do not. That sounds harsh, but I think it is true. AI will help people with existing skills. It will make productive people more productive. It will make knowledgeable people faster. It will help people who understand the fundamentals build, edit, automate, analyze, and create more efficiently. But it will not magically give people judgment.

I was talking this past week to a director at one of the big accounting firms. He has been using AI for about as long as I have, roughly four years. He started at the bottom of the company and worked his way up, so he has done the basic work by hand. He knows how the machinery works because he once had to turn the gears himself. He told me that their AI now handles much of the work he did when he first started. For example, many associates are expected to use AI to help with reconciliation reports. But when the reports are wrong, the associates often do not catch the mistake. He does. Because he has done the work manually, he can look at an output and immediately know when something is off. That is the part people keep missing. Skill is still required.

The developer behind me could use Claude effectively because he knew what he was trying to build. The accounting director could catch AI’s mistakes because he understood reconciliation before AI entered the room. When I use AI to build automated reports, I still need to know what the report is supposed to do. Otherwise, I am just politely asking a machine to produce nonsense in a more efficient font.

AI is like the new Google. When Google first entered classrooms, it changed how people found information. I remember that shift. Suddenly, answers were faster. But Google did not make everyone smarter. It helped people who knew what to search for, how to evaluate sources, and what to do with the information once they found it. AI is doing the same thing, only faster and with more confidence. Those who are efficient, skilled, and grounded in the fundamentals will gain the most from AI. Those who are not will get fluent-looking mistakes, automated shortcuts, and perhaps a very polished road to nowhere.

This is why I worry about students in high school and college. I do not think students should use AI to replace the act of learning. Teachers should also, as much as possible, try to prevent students from relying on AI before they understand the work themselves. I know that sounds old-fashioned. I am aware that every generation eventually becomes the people warning that the new tool will ruin the youth. Somewhere, someone once looked at a calculator and thought, “Well, civilization had a good run.”

But there is a real danger here. If students use AI before they develop their own skills, they will not become the people who benefit most from AI. They will become dependent on it. They will have the tool, but not the judgment. And in a capitalist world, that is not a great position to be in. Capitalism is not famous for gently protecting people who lack bargaining power. Before you outsource the thing, you need to learn the thing.

There is another problem, too, and it is harder to measure. AI can flatten the human soul. That sounds dramatic, but I mean it plainly. Human beings need purpose. We need to make things. We need to write words, draw pictures, cook meals, take photographs, tell stories, arrange furniture badly, rearrange it, and then pretend the second version was the plan all along.

Everyone is creative, even people who insist they are not. Some people are creative with words. Some with numbers. Some with recipes. Some with spreadsheets. Some with gardens. Some with jokes. Some with the exact placement of a throw pillow, which I do not personally understand, but am willing to respect. The point is that creativity is not only for artists. It is part of being human. And right now, people already spend so much of their lives consuming other people’s creativity through algorithms. We watch, scroll, react, compare, and absorb. AI may make this worse. There are already people asking AI to come up with the idea, write the essay, draw the picture, compose the song, generate the caption, and then tell them when to post it.

That should not be the goal. Use AI as an editor. Use it as an assistant. Use it as a curator, organizer, or second set of eyes. Use it to challenge your thinking, clean up your structure, or help you see what you missed. But do not let it become your ghostwriter, ghost painter, ghost thinker, or ghost soul.

Do not let a machine do all the making for you. Write the words. Take the picture. Draw the thing badly. Make the playlist. Cook the meal. Put something of yourself into the world, even if it is imperfect. Especially if it is imperfect.

AI can be a good tool. It has made editing available to more people, including people who could never afford the kind of support that famous authors have always had. That part is genuinely useful. But relying on it completely will make us smaller.

And then there is the money. If we become overly reliant on AI, someone will find a way to extract money from that reliance. The tools will become subscriptions. The subscriptions will become tiers. The tiers will become necessities. And the people who own the platforms will discover, with great sadness and a prepared statement from investor relations, that prices simply must go up.

When I say “poor people,” I mean almost all of us. I mean the 99%. If you do not own a private jet and a mansion, you are probably not on the winning side of this arrangement. You may own a house. You may own land. You may have a retirement account and a nice sectional. That can create the illusion of safety. But the people building these systems do not think of most of us as citizens, neighbors, or fully formed human beings. They think of us as users, consumers, data points, labor costs, risks, and claims to be denied.

AI will not only write emails and summarize meetings. It will increasingly help decide who gets approved, who gets denied, who gets hired, who gets flagged, who gets monitored, who gets care, and who gets told to call another number. And of course, this will all be sold to us as efficiency. That is the word we should probably fear most. Efficiency sounds so reasonable. Who could be against efficiency? Only a fool would oppose efficiency. But efficiency for whom? Efficiency at whose expense? Efficiency measured by what?

A system can be very efficient at denying people help. A company can be very efficient at reducing labor costs. A platform can be very efficient at keeping you addicted. A government can be very efficient at deciding that you are not worth the expense.

Picture some future version of Medicaid or Medicare relying heavily on AI systems to evaluate claims. The companies selling the systems will say they are saving taxpayer money. They will say they are protecting the program for people who truly deserve it. They will say the model is objective, as if objectivity is something you can purchase in a software package.

And then, one day, someone will discover that they can be born here, live in the Midwest, wear the flag proudly, vote correctly according to whoever is currently defining “correctly,” and still not be considered deserving enough when the algorithm says no. That is the thing about systems built to exclude. Eventually, they come for people who thought they were safely inside.

The information environment will get worse, too. A small number of powerful people and companies already control much of what we watch, read, and scroll through. The AI systems we use are also owned by powerful companies with their own incentives. Americans often imagine censorship as something that happens somewhere else, in countries we have been taught to pity or fear. But we have been living inside softer forms of control for a long time.

Your social media feed shows you things designed to keep you from leaving the app. News channels package reality in ways that keep you watching. Platforms do not need you to understand the world. They need you to stay engaged long enough to be monetized.

AI will make that easier. It is not that doctored images, propaganda, or manipulated narratives are new. Dictators have been altering reality for a very long time. Stalin did not need Photoshop to remove people from photographs. Hitler did not need a generative model to flood a country with lies. AI is not inventing manipulation. It is industrializing it. The wealthy and powerful will not need as many people to produce propaganda, distortion, or distraction. They will be able to automate it. At scale. With better grammar.

And then there is the environmental cost. AI is often presented as weightless. Digital. Clean. Floating somewhere in “the cloud,” which sounds peaceful until you remember that the cloud is mostly buildings, servers, electricity, heat, water, land, and shareholders. Data centers require enormous amounts of power and water. They need infrastructure, cooling systems, and energy. The companies building them will describe this as the cost of progress. They will say AI is good for humanity. They will say it will free up resources, improve productivity, and solve problems that previous technology helped create.

Maybe some of that will be true. But I worry that the public costs will be carried by the public, while the private profits remain very private. We have already seen what happens when wealthy industries are allowed to pollute air, water, and land in the name of growth. The damage does not fall evenly. It never does. The people with the fewest resources are usually the first to breathe the bad air, drink the bad water, and live near the consequences. All of this, so we can generate more content. More emails. More images. More summaries. More posts. More slop. More little bits of language and light to keep the machine full.

So as the man behind me talked about spending a large portion of his salary on Claude to move up in his company, I saw more than one ambitious developer trying to get ahead. I saw a small preview of the future.

A future where the skilled become more efficient, the unskilled become more dependent, creativity becomes outsourced, information becomes more manipulated, jobs become more precarious, public goods become more strained, and the wealthy discover yet another way to turn human anxiety into recurring revenue.

I wish I had a more cheerful conclusion. I would prefer to end with something balanced and hopeful, perhaps involving a sunset over the Rockies or the men behind me waking up with a renewed commitment to labor organizing. But I do not feel especially cheerful about AI.

I do not think the danger is that AI will kill us like in Terminator. That would at least be direct. Dramatic, yes, but efficient. No subscription tier required. The more likely future is slower and more humiliating. More claims denied. More jobs squeezed. More creativity outsourced. More water used. More heat generated. More money extracted. More people told that all of this is innovation.

Hopefully, I am wrong. I would like to be wrong. But sitting on that plane, listening to a man describe spending his own salary to keep up with the machine, I could not shake the feeling that the future had already boarded. And, of course, it had priority.