My son called me last night. Not the usual check-in. Not about classes or weekend plans. He had a problem.
"Dad, I ran out of credits."
He's been using Claude Code to build things, writing software, exploring ideas, learning by doing. He's caught the bug. And as a dad who's spent his career in technology, I couldn't be more thrilled. He'd been working within the constraints of a standard plan, and honestly, the constraints were good for him. He learned to be resourceful. But he hit the wall. Who am I to stop a kid from running agents in parallel. So I gave him my credit card and signed him up for a Max plan to unblock him.
I had the privilege of being able to do that. But that phone call stuck with me, because for many families that is probably hard to do. And I think that points to something much bigger than one kid running out of tokens.
Thirty College Tours
I have college-age children. Over the last three or four years, I've probably visited thirty campuses. If you've been on the college tour circuit, you know the script. It's remarkably consistent across schools.
At some point, the guide walks you into the library. They gesture at the stacks, the study rooms, the reading nooks. "And if we don't have the book you need, we're part of an inter-library loan network. We can get it shipped from another university." You nod politely.
Then they take you to the maker space. You see the 3D printers humming away, maybe a laser cutter, some soldering stations. "Students can come here anytime and bring their ideas to life!"
And I'm standing there thinking: you are fundamentally missing the point.
Knowledge in 2026 is not hardcover books, no matter how romantic they are. And making is not just 3D printing. Those things are great, and physical building still matters. But the knowledge that matters now also includes access to intelligence. To AI. To tokens. And the making that matters increasingly is what happens when a student sits down with Claude Code or Cursor and builds something real in an afternoon that would have taken a semester of coursework five years ago.
If one of the goals of higher education is to prepare students for gainful employment and creative contribution, then we need to give them access to these tools. They should be part of the infrastructure, budgeted the same way universities budget for Wi-Fi and lab equipment.
I pay an extraordinary amount of money in tuition. Don't even ask me the number. These schools are showing me their 3D printers and inter-library loan networks, and then I go home and pull out my credit card to buy my kids the tools that actually matter. Something is wrong.
I see this as three simultaneous failures:
The education system should be providing these tools the way it provides lab equipment, software licenses, library databases, studio space. AI access belongs in that same category. It's becoming the primary tool of intellectual work.
AI companies — OpenAI, Anthropic, Google — none of them have built the institutional access models that education demands. They offer consumer subscriptions and enterprise tiers, but where is the academic tier that lets a university provide every student with meaningful access? Inference isn't free, but this is a solvable problem if it's treated as a priority.
Public policy hasn't caught up. We subsidize internet access in schools. We fund lab infrastructure. We build libraries. But there is no equivalent program for ensuring that students have access to AI, which is quickly becoming as foundational to learning as the internet itself.
The Wealth Filter
I almost made this worse. There's a feature in Claude Code, /insights, that shows your AI usage patterns. When I first saw it, I was excited. If someone applying for a role at my company showed me that report, that would tell me a lot about whether they get this moment. A signal that they're actually building with AI, daily.
But if I make AI fluency a prerequisite, if I expect candidates to show me their power-user stats, I'm penalizing every talented person who simply couldn't afford the tools. That's not a meritocracy. That's a wealth filter wearing the clothes of a skills assessment.
The thing I wanted to use as a hiring signal becomes exclusionary without the access infrastructure to back it up. And this is already happening, whether hiring managers realize it or not. Every job posting that asks for "experience with AI tools" is implicitly asking: could you afford $20 to $200 a month for the subscriptions that matter?
The divide between those who have access to AI tools and those who don't is widening now, in dorm rooms across the country. Some students are building software, exploring research questions, developing fluency with the tools that will shape their careers. Others are locked out because their family can't absorb the cost.
If we don't address this quickly, we're widening the divide.
Where Is Our Andrew Carnegie?
This isn't a new problem. Libraries were once private. If you're ever in New York, visit the Morgan Library & Museum. It was J.P. Morgan's personal library, built in 1906. Illuminated manuscripts, rare first editions, centuries of human thought gathered under one roof. Built for one man.
Knowledge was expensive and concentrated among those who could afford it, until Andrew Carnegie, who knew this personally. He'd immigrated from Scotland at age 12, gone to work in the factories of Allegheny, Pennsylvania, and got access to books only because a retired businessman named Colonel Anderson lent them to local workers every Saturday. That experience changed his life.
Between 1883 and 1929, Carnegie funded over 2,500 free public libraries across the country, spending more than $50 million in early 1900s dollars. He didn't wait for the market or the government. But he didn't do it alone either. He required each community to fund ongoing operations. Philanthropic capital to build, local commitment to sustain.
"A library outranks any other one thing a community can do to benefit its people." — Andrew Carnegie
We need a Carnegie moment for AI.
What a New Library Could Look Like
Universities already have a model for this. They negotiate consortium licenses for academic journals and databases, bundling institutional buying power to give every student access to resources that would be unaffordable individually. That same infrastructure could work for AI.
A university negotiates with Anthropic, OpenAI, and Google for institutional access. Every enrolled student gets a meaningful monthly allocation of credits as part of their tuition, the way they get library access and Wi-Fi today. A real academic partnership, with real capacity behind it.
AI tokens have a real marginal cost every time they're used. You can't lend a token the way you lend a book. But that makes it more like a lab budget than a library shelf. Universities fund chemistry labs and computing clusters. AI access is the same category of investment.
Over time, the model will evolve. Some institutions will host open-source models on their own infrastructure. Some will blend commercial access with self-hosted alternatives. The technology will get cheaper. But we can't wait for "cheaper" to solve an access problem that's happening now.
I know this is messy. The technology is evolving fast and it's hard to build programs around a moving target. But Carnegie didn't wait for books to get cheaper.
Somewhere right now, a student is hitting the same wall my son hit last night. My son got unblocked with a credit card. Another student might not. One kid keeps building. The other stops. That's the gap we need to close.