The End of the Subsidy

As I write this, my work laptop is sitting next to me churning through a complex upgrade to an outdated codebase. The strategy is mine. The plan is mine. The execution is entirely AI. It’s been running for hours, working through problems I scoped and decisions I framed, and the output so far is genuinely good. This is not a demo. This is Tuesday.

A year ago that sentence would have been extraordinary. Now it’s just how I work, and that shift happened so fast that it’s worth pausing to ask why. Not why AI is capable of it, that part is obvious by now, but why it felt so easy to get here, why the friction was so low, and why we rarely stopped to ask why the magic was so cheap.

The answer is that someone else was paying for it.

Behind every interaction sits infrastructure, compute, and research spending at a scale beyond what most people can begin to comprehend. All of it absorbed so that the experience on our side felt close to free. Subscriptions blurred the edges. Generous limits hid the meter. Even the lowest tier plans gave back multiples more in tokens than what we were paying, and it felt exactly the way it was supposed to feel. It felt like abundance.

That generosity wasn’t accidental. It was an investment. The hardest part of AI adoption was never the technology. It was belief. If you can get someone to experience the upside without confronting the cost, to feel the output as disproportionate to the input, the argument is over. They don’t need convincing anymore. They’ve felt it. Our onboarding into AI was underwritten specifically so that we would see the light, and we did.

People explored freely because that was the whole point. They built things they wouldn’t have tried otherwise. They iterated without watching the meter. Companies encouraged it. The math didn’t need to close yet because the value was uncertain, the upside looked enormous, and the only rational move for everyone involved was to explore as broadly as possible.

Now the subsidy is being withdrawn and the underlying economics are surfacing. GitHub Copilot is moving toward explicit credit models. Anthropic is tightening what fits inside base plans versus what requires additional spend. These are not product tweaks. The system is correcting itself, aligning usage with cost now that the belief has been established.

Nobody serious is questioning whether AI produces value anymore. That laptop next to me is the answer to that question. What’s being asked now is what the value is worth when tokens are a line item instead of an invisible mechanism, when every interaction carries a price, when you’re the one paying.

For individuals the change is quiet. You start thinking before you run something. You tighten prompts. You reuse context instead of regenerating it. Iteration develops weight it didn’t have before. For businesses it’s more immediate, token spend starts looking like any other operational cost, and the same expectation applies: it has to produce more than it consumes.

That laptop is still running. The work it’s doing is real, the value is real, and starting tomorrow it’s going to cost me more than it did yesterday. That’s fine. It’s actually just business as usual. Every tool I’ve ever used professionally has required an investment of time and effort measured against outcomes, and the best ones have always paid for themselves. AI is not an exception to that, and what comes next is just the normal work of making good investments. The subsidized introductory offer helped enable me to explore this AI era as enthusiastically and deeply as I did, and to see the potential of what this new tool can offer. Now though, as ever, it’s up to me to ensure I’m investing my time and money wisely, not just spending it.