The New Hidden Cost Of Development

Kamil Rogozinski | | 5 min read

If you’re like me, you’ve noticed the surge in AI tools—Claude Code, Codex, OpenCode, and others. We live in a time where anyone can build software. Got an idea for a custom dashboard? Subscribe to a plan, guide the model, watch it create what you need. Hit your limit? Upgrade to the next tier and keep going.

It’s remarkable, really. The barrier to creating software has never been lower.

But I keep thinking about what we’re trading away.

I got into coding because it was a skill—something you could learn through effort alone. You could pay for a course or a mentor’s insight if you wanted, but you could also figure it out on your own. Documentation was free. Stack Overflow was free. The struggle itself was free, and that struggle was where the learning happened.

Now the path of least resistance runs through a subscription. And I’m not sure what that means for the future.

The Shifting Baseline

The financial barrier to entry used to be almost nothing. A computer, an internet connection, and time—hours spent debugging, reading documentation, building things that broke.

Now we’re substituting money for that time. The learning cost hasn’t disappeared; it’s transformed.

But it’s not just price. It’s expectations.

Job posts are starting to imply these tools. Teams are integrating them into workflows. Dev Twitter is full of “I built this in 30 minutes” demos where the unspoken footnote is: with a paid AI stack.

So the question shifts from “Should I use these tools?” to something harder: “If I don’t, am I falling behind?”

Plans that started at $20 added a $100 tier. Some have climbed to $200. And if the “standard developer setup” quietly becomes IDE, GitHub, cloud hosting, plus three or four AI subscriptions—then being unable to subscribe doesn’t just mean fewer conveniences. It means fewer reps, less output, a weaker portfolio.

The Apple Parallel

I think about how Apple products evolved. Prices went up as demand went up. Some of that was justified—better components, more capability. Some of it was simply what the market would bear.

AI coding tools feel like they’re on a similar trajectory, except more aggressive—because these aren’t one-time purchases. They’re subscriptions. They compound.

And unlike a phone, this isn’t a luxury gadget. It’s increasingly tied to work, learning, and momentum. As dependency grows, so does pricing power.

The Deeper Trade-off

A question from a recent Anthropic study stuck with me: Does AI provide a shortcut to both skill development and increased efficiency? Or do productivity increases from AI assistance undermine skill development?

That tension is what prompted this post.

The study found that developers using AI assistance scored 17% lower on comprehension tests—nearly two letter grades—compared to those who coded by hand. The largest gap was in debugging: understanding when code is wrong and why it fails.

There’s a real difference between “I fixed it, and I understand why it broke” and “It works now, and I’m not sure what changed.” Both ship code. Only one builds confidence.

I’m not romanticizing suffering. Getting stuck for days isn’t noble—it’s just slow. AI can be a tutor, a debugger, a second brain. But the study points to something important: how you use these tools matters. Participants who asked follow-up questions, requested explanations, and used AI to check their understanding retained far more than those who simply delegated the work.

Use these tools too little, and you progress slowly. Use them too heavily without engaging, and you might progress shallowly. That tension isn’t going away.

Who Gets Left Behind?

The hardest question isn’t personal—it’s structural.

For those who can justify or afford $200 or $500 a month subscriptions we live in great times. But what about the teenager learning to code in their bedroom? The career-changer studying at night? The person in a country where $20 a month is already significant?

For years, programming was one of the most accessible skills escalators available. The internet made it possible to grind your way into competence with enough time and patience.

If AI becomes the default way software gets built—and if the best AI is something you rent—then we’re quietly rebuilding a gate that used to be smaller. Not by making learning impossible, but by making it slower and lonelier for those who can’t subscribe.

I Don’t Have Answers

I’m not writing this from the outside. I use these tools daily, and they’re genuinely transformative—explaining unfamiliar codebases, catching bugs I’d miss, helping me prototype faster. This isn’t a critique from someone who’s opted out.

But I’m watching a skill I valued—something learnable through effort—become increasingly mediated by what you can afford. The gap between those who can subscribe and those who can’t starts to mirror the gap between those who can build and those who can’t.

Then again, maybe that’s only half the picture. These same tools could lower the barrier for people who never saw coding as an option—those who found the learning curve too steep or the time investment too large. They might bring ideas and perspectives we’d never have seen otherwise. The democratization could cut both ways. There’s real reason for hope. Kimi K2.5 just dropped with benchmarks rivaling closed models. If that trajectory continues, capable models running locally on your own hardware—no subscription required—might become the norm rather than the exception. Competition could drive prices down. Open-weight development could close the gap entirely.

Maybe the fundamentals will remain just as learnable without paying anything at all.

I hope so. Right now, I’m still watching the trajectory.