AI-LMC Phase 00 — I just wanted to make easy money with AI, so obviously I'm now building a legacy app from scratch 🐼 (Part 1)

I want to make money with AI. Ideally, easy money. To put it more precisely: I want to hand as much as possible to AI, and do as little as possible myself. 🐼 "Is there a convenient method like that somewhere?" So I asked the AI first. Please, AI, let me be lazy When I scroll through social media, I keep seeing people overseas post things like: "I made money by having AI do this." "I built this system and it just runs on its own." Must be nice. I want that too. 🐼 But some of those services can't be used from Japan. For example, Polymarket, the prediction market you see a lot overseas, currently lists Japan among its blocked countries. And there are plenty of others. You find some clever scheme going viral abroad, and then: "Great, let's just do that in Japan!" …turns out you can't. So I asked my GPT and my Claude. 🐼 "I want to hand things off to you two and get to a point where I make money without much effort." We talked about a lot of things. A lot of ideas came up. But in the end, GPT and Claude both circled back to roughly the same place. 🤖 "The fastest route is to use the eleven-plus years of engineering experience you already have." 🐼 "…So we're back to that?" "Use your eleven and a half years," they said I've been an engineer in Japan for about eleven and a half years. Mostly C# and .NET. But when I look over the résumé of those eleven years, it's almost entirely internal systems and B2B line-of-business applications. I haven't built consumer web services. I haven't maintained a well-known open-source project. Windows applications used inside companies. Point-of-sale systems. Systems built to make some back-office process less painful. That's what I've been making, the whole time. And in this kind of work, some projects restrict the use of generative AI entirely, because the systems handle confidential and personal data. Naturally, none of that source code is on GitHub either. So when I go looking through GitHub issues and public job listings thinking: 🐼 "Maybe I can get AI to do this part and take it easy" …the kind of work I've actually done for eleven years doesn't show up. Of course it doesn't. They're internal systems. 🐼 "So what am I supposed to do with my eleven years in the age of AI?" GPT and Claude gave more or less the same answer to that one too. Legacy Modernization. Apparently Japan still has a lot of legacy I looked into it. According to a 2025 report from Japan's Digital Agency, based on the summary report of the Legacy System Modernization Committee: 61% of user companies still have legacy systems. Among large enterprises: 74%. 🐼 "That's a lot." My eleven years of line-of-business experience applies. C# and .NET apply. And there's still a mountain of legacy systems out there. Right then. 🐼 "So I do Legacy Modernization and I get paid?" 🤖 "No." 🐼 "…" But isn't AI going to do that anyway? .NET 10 was released in November 2025. Thanks as always, Microsoft. 🐼 And .NET 8 reaches end of support on November 11, 2026. Which means upgrades like: .NET 8 → .NET 10 are exactly what people are doing right now. Microsoft also already has GitHub Copilot modernization. It analyzes a .NET project, produces an upgrade plan, modifies the code, and verifies the build and tests. There is an official scenario for exactly this: "upgrade this solution to .NET 10." 🐼 "…" 🐼 "So AI can already do it." …Really? And here's where something caught. Really? I mean, sure. Rewriting the target framework. Updating NuGet packages. Fixing compile errors. AI is going to keep getting better at all of that. But. The legacy line-of-business systems I've actually seen — were they ever that tidy? For example. "Only A-san understands this part." "Where's A-san?" "Quit last year." Or. "The spec says this, but that's not how it actually runs." "Which one is correct?" "You'd have to ask the people who use it." Or. "Why is this one number corrected by hand?" "It's always been done that way." "Who decided that?" "No idea." Or. "We could probably delete this, right?" "No — apparently some department has a problem at month-end if you remove it." "Which department?" "Accounting, probably." "Probably?" In a legacy business system, reading the source code is not enough. It isn't in the documentation. The spec is out of date. Only one person knew. And that person is gone. So an engineer goes around asking. "What is this process actually for?" "How am I supposed to decide this value?" "Is this workflow still in use?" I've watched that happen many times. Which is when it occurred to me. 🐼 "Can AI figure this out too?" If you analyze the source code, can you really recover the things that used to require asking someone? Even human engineers couldn't. They went looking for the person who knew, asked the question, failed to get an answer, and decided for themselves in the end. Can AI really work all of that out? 🤖 "That question itself is interesting." 🐼 "Hm?" Part 2 continues.