Mike Swarm

Made this with Fable 5.5 max effort. API Cost $68

You are now unstoppable! Don’t let the engineers tell you otherwise!

This is false, you can maintain millions of lines of code if the architecture is done right, don’t listen to these gatekeepers who are scared to lose their jobs.

This is a life or death move and major gamble with mind boggling context hidden within that few grasp. Cisco believes they are finished without it, and few people can get the job done. What a time to be alive.

Stop using higgsfield!! They are the worst in the industry!

Almost a 100 billion tokens a month with Claude code, mainly opus 5.5 Getting so close to completing an operating system for my business. Reveal coming soon!

Vibecoders are going to devour the gaming industry Everyone and their mother are currently developing games, in a year or two there will be thousands of games being dumped on the market every week.

There is 0 chance OpenAI’s dot is using Astra. Did they silently downgrade it? This is worse than the car wash walk.

I’m officially done with OpenAI OpenAI seems to have changed how “Dots” work, it used to do most of the work within its own virtual PC, now it delegates the work to other codex sessions that use up your limits (same exact type of work it used to do on its own VM) I’m officially done with OpenAI, almost everything is bait and switch with them. There is absolutely no value left even if they release better models, Anthorpic is way ahead.

If you use Claude code and have a Linux box you need to install TMUX, it allows you to organize sessions, and connect from anywhere, and if you lose your ssh connection or your internet times out you don’t lose your work, it’s a lifesaver!!! I can easily handle 8+ sessions and several Claude accounts that way.

We all have that one friend that thinks the new Tesla roadster is a gimmick. Send him this 😂

“I built this with AI in 20 minutes.” Cool. Who’s still using it 20 days later? Spend enough time in the AI bubble and it starts to feel like you’ve already missed the opportunity. Another launch. Another demo. Another person announcing that they’ve built an entire company over the weekend. But building something and building something people actually need are two very different things. Look at Karpathy’s rough AI funnel: around 6 billion people have barely interacted with LLMs, 1–2 billion use chatbot-style products casually, and roughly 20 million use frontier models for technical work such as coding. These are his back-of-the-envelope estimates, not a global census. That 20 million is where I think the more interesting funnel begins. Because 20 million advanced users does not mean 20 million people building useful products. It isn’t even a count of founders. My suspicion is that a substantial chunk of the building activity is producing useless shit: broken games, copycat tools, abandoned prototypes, and AI wrappers with no clear reason for anyone to choose them. Projects built around “look what AI can generate,” rather than “here’s a problem someone desperately needs solved.” And before someone misses the point: experimenting is how you learn. A rough game can be a useful learning project. A wrapper can be a valuable business. The problem isn’t the format. It’s confusing the ability to make something with evidence that anyone wants it. So let’s extend the funnel with an explicitly hypothetical scenario. Suppose 10–25% of those 20 million are consistently turning that capability into genuinely useful products, tools, research, or working systems. That leaves 2–5 million people. Now suppose 0.5–1% of that smaller group eventually achieves a major outcome: substantial adoption, a durable business, or a meaningful breakthrough. That leaves 10,000–50,000 people. Tens of thousands, on a planet with more than 8 billion people. Those last two steps are assumptions, not measured totals or a forecast. But they illustrate the distinction that matters: Using AI, generating a product, solving a real problem, and building something highly successful are not the same achievement. A polished demo doesn’t prove demand. A launch doesn’t prove retention. And a thousand people commenting “insane” doesn’t prove that anyone will pay. The useful questions are less glamorous. Does it work outside the demo? Does it save someone meaningful time or money? Do people come back without being chased? Would anyone care if it disappeared? That’s what I mean by doing something serious with AI. Not necessarily inventing a new model or building a billion-dollar company. A tool that reliably fixes an expensive problem for a specific group of people counts. If that’s what you’re working on; understanding the problem, testing with real users, fixing what breaks, and continuing after the excitement wears off; don’t measure your progress against the volume of launches in your feed. Your competition isn’t everyone with an AI subscription. It’s the people serving the same customers and solving the same problem better than you. Being part of a small group doesn’t guarantee success. But a crowded feed doesn’t prove that your opportunity is gone, either. Don’t aim to build something people are impressed AI could make. Build something they would be pissed off to lose.

6.5 bilion tokens in 24H with Opus 5.5