Software Abundance Is Here. The Bar Just Got Raised.

It's 12:30 on a Saturday morning. My wife wants to go to sleep, and I'm showing her a little rocket ship on a dashboard that tells me how hard my AI agents are pushing the servers. She's being a good sport about it. I'm vibrating with energy, because that silly little dashboard is the clearest proof I've found that software abundance is here, and that the bar for every business on the planet just got raised.

Here's my thesis in one line. Abundance hit software first, it hit this month, and everything else in your business is next. I've got the receipts from my own fleet to back it up, and I'm going to show you all of them.

Software abundance started with a pet project on my desk

The dashboard started as a pet project. I run a fleet of AI agents across a bunch of Claude and OpenAI Codex accounts, and I wanted to make sure every account burned its tokens perfectly, with predictions so I never ran dry at the wrong moment. Then it started load balancing between Claude and GPT on its own, switching the models my agents use based on burn pacing.

Then I got carried away. I added a rocket ship so I could see how hard I was pushing the fleet. Then disk, CPU and RAM gauges. Then light mode, dark mode, sound effects, hidden Easter eggs, a high score, and a space trip to Mars. There's an accelerated burn mode that force-transfers load between accounts, and a "max concurrent" strategy that spreads sessions out so I don't get rate limited, because you only get about 8 to 10 connections per CLI account before the provider pushes back.

Under the hood there are layers and layers of predictive math. And it all runs on plain CPU. It doesn't burn a single token.

Here's the part that got me out of bed. I built it in five-minute bursts over five months, maybe 40 or 50 hours total. With today's models, I could rebuild the whole thing in an afternoon.

I could rebuild that whole thing in two hours for like $10. And that's the point here. Abundance is here for software and abundance is coming for the rest of the world. Mike Schwarz, Founder & CEO, MyZone AI

And that dashboard is one module on one page of my portal. I'd guess it's about 1% of what my agents and I have built this year. So what happens when the other 99% gets the same treatment?

Opus 5.5 made software abundance show up in my own data

On September 22, Anthropic shipped Claude Opus 5.5. TechCrunch called it Fable-level performance at a lower price, and in my testing it feels at least that good, often better, and noticeably faster. The list price is $4 per million input tokens and $20 per million output, down from $5 and $25 for Opus 5, and Anthropic says typical workloads cost 40% less once you add in the efficiency gains.

Compare that with Fable 5.1 at $10 input and $50 output. That's 60% cheaper on the rate card alone, and my napkin math on effective cost landed closer to 70%. Then I pulled the real numbers from my fleet, and they're even better than my napkin.

For most of the last month my fleet ran two to three billion tokens a day. In the first three full days after Opus 5.5 launched, it averaged 6.4 billion, and it hit 8.3 billion on September 25. I didn't hire anyone. I just stopped rationing.

The reason I stopped rationing is the next chart. This is what each model actually costs me per billion tokens once caching, reasoning and real agent workloads are all mixed in.

Now put volume and price together. On September 7, one of my heaviest days, the fleet ran 4.6 billion tokens. On September 25 it ran 8.3 billion, and spent less money doing it.

That's what software abundance looks like up close. Output goes up. Cost goes down. Same week.

Tokens are the new capital, and the compute race proves it

Zoom out and you see why this keeps happening. The labs are in an all-out race for compute, and the numbers are wild. xAI's first Colossus data center made headlines with 100,000 GPUs, and Anthropic now rents all of Colossus 1 for $1.25 billion a month, more than 300 megawatts of capacity.

Elon has rolled xAI into SpaceX and just keeps building. He told employees he'll grow data center capacity 7x to 10 gigawatts by late 2027. For perspective, where he is today, roughly 1.4 gigawatts, is enough electricity to power about a million homes.

And compute isn't only getting bigger. It's getting faster. NVIDIA paid about $20 billion to license Groq's inference technology and now claims up to 35x more inference throughput per megawatt on its new LPU chips. Faster tokens mean agents that think and act in something closer to real time.

Meanwhile China keeps closing the gap with far less compute. Part of how is distillation: hammering a smarter model with millions of questions and training on the answers. Anthropic says seven China-based labs ran more than 190 million queries through fake Claude accounts between May and July. The American frontier labs are, in effect, giving birth to super low-cost offshore models.

One of those, GLM-5.3, is now open-weight, and you can run it locally on a well-specced Mac Studio. My prediction: in early 2027 a lot of compute moves from the cloud back into the home and the office, and tokens get close to free for anyone who owns the hardware.

Having access to AI tokens is like having access to capital, right? You need to think about it that way. Mike Schwarz, Founder & CEO, MyZone AI

So ask yourself: who in your company is in charge of locking down compute? If the answer is "nobody," you have a capital problem you haven't named yet.

AI is raising the bar for every business, not just software

Software is simply where abundance shows up first, because software is made of the same stuff AI makes: text, logic and decisions. As the price of software drops toward the price of the electricity that runs it, everything I just showed you gets 10x cheaper and 10x faster again. My bet is six to 12 months.

Now picture your competitor. They automate their operations and drop operating expenses 20% to 30%. Where does that leave you on price? And if you're in professional services and someone drops their costs 90% or more, what happens to your whole industry?

This isn't hypothetical. AI-native law firms are already pairing agents with human lawyers who review and sign off at the end. The UK's Lawhive raised $60 million on exactly that model. Agents do the heavy lifting, a lawyer puts their stamp on it, and the client pays a fraction of the old price in days instead of weeks.

It's like if you're in construction and you're building houses, like now you are building skyscrapers by yourself. You do not need 500 people. You do not need financing. Mike Schwarz, Founder & CEO, MyZone AI

And this skyscraper grows wider and taller every day. If your competitor is selling condos in it at a tenth of your cost, how are you going to compete by building houses one at a time?

You don't need to be a coder to ride software abundance

You might be thinking I'm some kind of lifelong hacker. I'm not. I started coding in January, nine months ago. What I do have is decades of running software companies, so I understand the architecture: how information flows through a business, how org charts and processes fit together, where the bottlenecks hide.

That's the skill that matters now. If you can see how a business should be structured, you can pick up the software part on the fly in a few months. A little ADHD helps too.

The other big shift is how you build. We no longer have to plan everything perfectly up front, because rebuilding is cheap. You can build Salesforce tomorrow, realize you did it wrong, and rebuild the whole thing.

Build your prototype, get it to work, it doesn't matter how shitty it is and then rebuild it. Mike Schwarz, Founder & CEO, MyZone AI

This week I started rebuilding my old nightlife website off the side of my desk: a social network, registration, marketing, community, photo gallery uploads, member sharing, photo tagging, around 50 features in total. It's up, hosted, secure and private. I've spent about two hours on it. Next up, live GPS updates so people can follow the party bus across Vancouver and tag their friends in real time.

The same thing is coming for marketing. A good-looking website isn't enough anymore. Your site should adapt in real time to the weather, the news and each visitor's history, and I've started a little gaming studio of agents that I'm connecting to Unreal Engine and ElevenLabs, so you can walk into a 3D world and talk to them. One person can build that off the side of a desk now. What could your team build?

Small is going to take down the big

When one person can 100x their output (I have, and I know plenty of people who have), a team of five starts to feel like a team of 500. My fleet runs hundreds of agents and hundreds of skills. It's basically an app store of automations I built myself. A small team with hardly any employees is shipping agent systems that big companies are only now announcing as version one.

That's why this isn't the time to bloat your team. The smaller you are, the faster and easier it is to change. And the gap between fast movers and everyone else doesn't stay the same size. It compounds.

Look at that chart again. Same speed. Same tools. The only difference is six months, and after two years the gap is 16 times bigger than where they started. That's the whole China-vs-US compute race in one picture, and it's your market too.

This is your final warning, and I mean it this time

I've been running full speed for four years and I still feel underprepared for what's next. I read Abundance 20 years ago, and I've been a futurist since I was 14, building video games on old Macs. I've preached this in my AI wake-up call for years. This is it. It's real.

So here's the version I'd tell you over a beer.

The beer version Like you're either in this or you're out. You know, if you're not in it and you're not going for it, you better sell your business or bring in a leadership team, a management team that's going to take over for you. Mike Schwarz, Founder & CEO, MyZone AI

If you're in, here's what that means this quarter. You need an autonomous fleet: agents built on agents, wired into real workflows. Start trading labor for tokens in the parts of your business that are pure process, and move your people up into the strategic roles only humans can fill.

• Lock down compute. Treat token access like a line of credit. Know your accounts, your limits and your burn rate.
• Automate one full workflow end to end. Not a chatbot. A process that used to take a person a day. My agent orchestration breakdown shows the architecture.
• Prototype, then rebuild. Ship something rough this month and let the next model clean it up.
• Stay small on purpose. Hire for systems thinking, not headcount. Here's how small teams run on agents.

And be ready for the ride. In my experience it goes exactly like this:

Are you ready to march forward five steps and fall four steps back over and over, get beat down with glimpses of brilliance and moments of frustration as you plod forward into a world of uncertainty? Mike Schwarz, Founder & CEO, MyZone AI

That's why I think this is the era of the ADHD-powered entrepreneur, the systems thinker who loves risk, loves learning and can change direction overnight. That person is going to gobble up markets. If you want the bigger economic picture behind all this, read why the abundance curve is bending.

There'll be no catching those Opus 6 and Opus 7 and Opus 8 powered companies. You just won't catch them. Mike Schwarz, Founder & CEO, MyZone AI

Software abundance is already here, and the rest of your business is next. Your competitors can already build skyscrapers by themselves, so what are you going to build with the next six months?