The Golden Age

Andy Brice asks, “Is the golden age of Indie software over?” He thinks it may well be ending, crushed between Google-driven enshitification, AI slop, and fraud. The decay of the open web poses a terrible threat to small internet businesses of all kinds. Certainly, most of the artisanal software crowd are experiencing hard times.

It’s worth noting, by the way, that two of our regulars have been living through war the last two years. Bomb shelters haven’t been part of our everyday plans for a long time. I suspect, though, that war is going to be a more frequent visitor in the coming years.

I think Andy Brice is too pessimistic, and that the best is yet to come, assuming the end of the world doesn’t arrive first.

First, the core problem is that we have only three sources of personal computers: Apple, Google, and Microsoft. Those three companies have worked to destroy artisanal software for more than a generation in order to secure more complete monopolies over the world of computing. They drive out competition and innovation by giving away software bundled with their hardware, and pretend to open up third-party opportunities while ensuring that those opportunities are only attractive to ignorant judges. But this cannot succeed long term because, in the long term, each customer wants to do their best work. If buying a better tool gets that novel finished or closes the next deal faster, you’re going to buy that tool.

This is made worse because Google is using its ad monopoly to build a software monopoly, and the road to that second monopoly requires treating independent software creators as equivalent to pornographers and online casinos. So, even if AdWords still worked, we are not permitted to buy them. It’s clearly an antitrust violation but no one has enforced those in ages aside from occasional big mergers. This won’t work in the long run, but it may be a near-run thing. I’m hoping my Senator and my Congreswoman may notice.

The pandemic still casts a shadow over software because it’s led lots of people to reassess their work, and led some people to just not care very much about their work. There’s still a lot of depression in the air. A lot of people get high on resentment; why work when you can have fun owning the libs and making jokes about trans people? This won’t last forever, but it might take a generation for the smoke to clear. (The last time it happened, the smoke cleared just in time for WW2. Just saying: keep smiling through just as you always do, and gather the rosebuds because they are not long, the days of wine and roses.)

AI will make a huge difference, but not in the way either the pundits or the Hacker News crowd seem to expect. Bob Martin has been crusading against vibe coding and he is not wrong: for code, Claude is wildly overconfident and has lousy taste. There’s apparently a plan at Microsoft to recode everything in Rust with a quota of a million lines of C++ per engineer per month. I’m sure that’s going to be peachy. And yes, some pointy-headed managers will stage big layoffs for the thrill of the thing, and hope to blame someone else for the consequences.

AI is bad at making stuff, and amazing at helping you make stuff. You don’t ask Gemini for the state of the tide, not if that matters to you, any more than you ask your eight year old kid. Sure, Bobby’s usually right, but if you don’t want to go aground, you’d better check. But AI is terrific at explaining how tides work and how we came to know, and AI is pretty good at telling you whom to ask. Even if the AI is wrong, you’ll find out right away; if it’s right, you’ll know the tides.

Much of the time, today’s AI is like a smart, unmotivated student: it knows a lot of stuff, but really has no idea what to do with that knowledge or why you would want do anything with knowledge in the first place. Remember: that’s not worthless!

• From time to time, the compiler gives you error messages that make no sense, either because you’ve hit some weird edge case in C++ or because you’re missing some typo. These consume a lot of time, even for very experienced programmers. Not anymore: Claude has gotten every one of these on the first try.
• Some necessary chores in software design are numbingly dull. Concurrency is like that: you need to work through long inference chains to assess the simplest change, and programming feels like one of those horrible chess positions where every piece is locked in place and protecting three other pieces. I get bleary, but Claude doesn’t mind at all.
• Same thing with code analysis: the compiler gives you a ten-deep counterfactual assertion that some pointer might be nil and crash. I stare and stare and cannot see how it could happen. Claude chews for a couple of minutes and explains and that’s that.
• Same thing with historical background. I’ve been thinking about writing a little piece about Judah Löw (1524-1609), the fellow in Prague who built a golem . Did he have a family? (Yep: wife Perel and seven kids—all of whom survived childhood). Where did he live? (Široká Street) Do we know anything about his wife? (Quite a lot!) Sure, I could dig this out the Harvard Libraries, and I probably will do it myself, but this isn’t bad work on a day when the library is closed. You can do it, too, even if you don’t have access to a good library. (Also, Claude can read German and probably Yiddish and Hebrew; I cannot.)

The theme for this year’s festival of artisanal software is artisanal intelligence. Don’t ask the AI to do your work; use the AI to help make your work better. Let it get you unstuck when you hit a roadblock. Let it work through the tedious chores.