Your $200 AI plan is doing $20,000 of work. It won't last.

I watched a single terminal session burn through $200 of AI tokens in two hours.

That was on an enterprise account, the metered, pay-per-token way companies buy AI, at a company I was building for. And I wasn't being careless. I had the full token-discipline setup: trimmed context, lean prompts, tooling chosen specifically to keep usage down. One developer. One session. Two hours. Two hundred dollars.

Now the part worth sitting with: $200 is also what those same AI companies charge me, as an individual, for an entire month of the max plan I run my studio on.

Two price tags on the same intelligence

Every major AI company now sells the same models two different ways.

Enterprises buy tokens: consumption-based pricing, billed per request, scaling with every prompt, every file read, every step an agent takes. Individuals buy a subscription, and the industry has converged on roughly $20 a month for a standard plan and $100 to $200 for the "max" tiers built for people who run AI coding agents all day.

I've used both, heavily. Metered at enterprise token rates, my own usage (full-time agentic coding, every working day) would easily bill past $20,000 a month. That's not a thought experiment; I've watched what two hours costs when it's billed per token. And it isn't just me: Gartner has tracked developer AI bills leaping from $20 or $100 a month to $2,000 or $5,000, with extreme cases hitting $20,000 in token charges. It now predicts that by 2028, AI coding costs will overtake the average developer's salary.

Two hundred versus twenty thousand, for the same models doing the same work. That gap is the most important number in software right now, and almost nobody is pricing it in.

Why would they let us have this?

I'm not going to pretend the AI companies are running a charity. Flat-rate consumer plans are a land grab: they build habits, lock in ecosystems, and generate the usage that makes the products better. Most subscribers use a fraction of what they pay for, and those light users subsidize heavy ones like me. And venture capital has been happy to fund growth over margins.

But that's exactly what makes this moment unusual. These companies could live without my $200. Next to their enterprise contracts, individual max plans are a rounding error, marketing with a payment attached. For a solo developer, though, that rounding error is the best deal in the history of the tools trade: enterprise-scale intelligence at consumer-scale prices. A genuine win on both sides.

For now.

The limits are already tightening

If you use these tools seriously, you've felt it. Usage caps arrive sooner than they used to. Weekly ceilings appeared where there were none. "Unlimited" quietly grew asterisks. Tools across the industry have been shifting from flat seats to consumption-based billing, and developers opening their first metered bills are discovering what enterprises always knew: token costs add up terrifyingly fast.

I've been saying this for a long time, and I'll put it in writing: we are living in the cheapest era of AI there will ever be.

To be precise about it: running yesterday's models keeps getting cheaper. Inference costs on older models have collapsed. But nobody building seriously wants yesterday's model. The price of the frontier, the newest and most capable models, hasn't followed it down, and agentic workflows multiply token consumption by orders of magnitude. Cheaper per token, times vastly more tokens, on the model you actually want: the bill goes up. The floor keeps falling while the ceiling, the part you live in, keeps rising.

And if the AI bubble bursts?

Half the industry is arguing about whether this is an AI bubble. For pricing, the argument doesn't matter, because every road ends at the same place. If the bubble bursts, the subsidies end, because the venture money funding below-cost inference dries up. If the market simply matures, the subsidies also end, because companies eventually have to show margins instead of growth. Either way, the era of selling dollars of compute for cents of subscription closes. Prices normalize up. Limits normalize down.

Nobody knows the date. Everyone can read the direction.

What a solo builder can do before it closes

The other half of this story is what the golden age makes possible.

The vibe-coding boom has non-developers shipping real products, and the solo-business wave keeps growing. But the deeper shift is what happens when actual engineering discipline meets a $200 max plan: a single senior developer is now, functionally, an engineering department. Architecture, implementation, tests, infrastructure, and review, running in parallel, all day, for a flat fee.

Anyone can make a website. That's been true for a while. What's new is that one person can now build the system: authentication, data, AI, compliance, cloud infrastructure, the whole stack. In a month of full-time, disciplined, AI-accelerated work, a solo builder can ship what an enterprise would have paid an agency six or seven figures to deliver.

I know, because that's how we work.

Our studio is built on this arbitrage, and we pass it on

This is the honest economics of Make My Web: we build production-grade systems using the same golden-age leverage this post describes, and we price them accordingly. The same build, done inside a big consultancy, would carry the tens-of-thousands-a-month token bills and the headcount. We deliver it for a fraction, not by cutting corners, but by standing on the right side of the biggest pricing gap in software. Coda, the HIPAA-compliant AI medical-coding platform we designed and built end to end, was built exactly this way.

So here's the takeaway, the same one I give friends: if you've been putting off building your system, build it while intelligence is cheap. The window is wide open today. It will not stay open.

Tell us what you want built. We'll tell you how we'd build it, and what it costs while the golden age lasts.