I recorded the raw, uncut version of this as a video on Tuesday night. Watch it on YouTube.
Grok 4.7 dropped Monday night. On Tuesday, OpenAI shipped GPT-6 Sol and Anthropic shipped Claude Opus 5.5. That’s three frontier models in roughly 24 hours, and all three are the mid-size working models you’d actually use every day.
I spent all of Tuesday burning usage. I didn’t know if the plans would bank my leftover usage or wipe it on reset, so I coded like someone told me the internet was closing at midnight. Both resets banked. Two big plans, two fresh buckets, and I still published an article in between.
The number I keep thinking about has nothing to do with benchmarks. As a company we spend over $1,000 a month on AI, and I burn about half of that myself. Our team has been shrinking while our output goes through the roof.
Which brings up the uncomfortable question under all of it. When everything you need to do gets done at the speed of thought, what are you for?
The $84,000 website
We used to charge $84,000 for a website. Believe it or not, that was a real price.
That same site now gets one-shotted in about two hours with a couple hundred dollars of inference. You point the AI at a site you like, give it your content, and tell it to build and publish. What comes out often looks better than what a professional designer handed over.
That business is evaporating. The labor got compressed into a line item on the inference bill.
Your ideas are the bottleneck
You can build anything and sell anything right now, and plenty of people are still broke. The old excuse was that you couldn’t do the work. You don’t have to do the work anymore.
So if it isn’t working, your ideas probably suck. Suck it up, buttercup, and come up with better ones. Burning through bad ideas has never been cheaper, and every one you burn teaches you something about the world you’re trying to build around yourself.
Knowing things pays better now
AI makes knowledge cheap to reproduce. It also pays out hardest to the people who know what to ask for. If you understand an industry, your prompts are sharper, and the returns stack on top of each other.
I see it most in people whose brains connect weird dots. The neurodivergent folks who link two unrelated ideas and find value in the gap are getting a much bigger return from this technology than everyone else. It feels a bit like getting superpowers, minus the cape.
What’s left for humans
If compute handles execution, I see two jobs left.
The first is trust. The internet is full of scams, and cheap AI made faking things easier. People still buy from people they trust, and trust still moves the slow way, through relationships, conversations, and showing up over and over.
The second is connecting the dots. Deciding what to build, for whom, and why. The AI will execute anything you give it. Somebody still has to point it somewhere worth going.
That’s why I think anyone willing to get in front of a camera and talk is part of the future. Show up consistently enough that people get to know you, and let the polish come later. You, or someone with an AI, will build everything else.
It’s also why my content keeps getting better, even on days I spend mostly coding. Some of it is AI-assisted, and it’s the best stuff I’ve made, because I did the thinking before I pressed the button.
Compute is becoming the electric bill
Watch the prices. Opus 5.5, GPT-6 Sol, and Grok 4.7 all came in dramatically cheaper, because Chinese open-source models have been price-dumping. Last generation’s frontier is now the middle tier, and it’s priced like one. More intelligence for fewer dollars.
I think compute ends up as a utility you pay monthly like power, and I’m rooting for it. I’m a big open-source guy. There’s a little cult forming around Claude right now, the “Claude, our lord and savior” crowd, and I don’t want any single lab holding that much sway over how everybody thinks and works. Open models keep the prices honest and the power spread out.
The arbitrage window is still open
Most people haven’t noticed any of this.
On Twitter (fine, X), my small circle is moving fast. People are raising billions, and I have friends there with a $23 million Series A. Then I open Instagram and AI barely registers. Somebody mentions Claude in passing and the conversation moves on.
That gap is where the money is. Arbitrage has always been the foundation of business, and my guess is we have something like ten years before the market fully catches up. There will always be some room left, but the wide-open version has a shelf life.
What I did with my extra tokens
Since I had the credits, I spent them on something we needed. We built our own multiplayer harness, a cloud setup that pools the team’s inference into one bucket. Nobody has to juggle separate ChatGPT and Claude accounts, we control spend in one place, and we talk to it inside our own chatbot. We might end up selling it.
Credit where it’s due: Bryan McAnulty, founder of Heights Platform, told me they’d built a multiplayer version of a tool I only had running locally. I said, yep, we need that. We trade ideas like that all the time, the same way I’m trading them with you here.
So what’s your purpose?
Let the machines take the work nobody wanted anyway. Your job is to be someone people trust, with ideas worth spending compute on.
This week, pick one idea and burn tokens on it until it either comes alive or dies. Then go have one real conversation with someone who could use what you know.
Let the AI flow.
What do you mean by "arbitrage?"