It’s around this time of year that Santa makes his first appearance on my blog. Want to get out in front of the others. #

My first job after college was on the 39th floor of the Empire State Building. I wrote software for our customers, in BASIC, on a teletype in our office, talking to our mainframe in New Jersey. It seemed magical at the time. When I hit a bug, a mystery — you’d try this and that, and not get anywhere on something that seemed so simple, you could stare at the problem for an hour, trying desperate things, getting no closer to the fix. One night, I was the only person in the office, I thought they should not put programmers in the Empire State Building with windows that open. I’m having the same feeling with Claude. I know when this project is done, I will have Frontier, and I know what that’s like. If I didn’t so totally look forward to that I would never try to finish the project, I hate this mode of working with the bot. Ugh. #

Just said this to Claude: “This is a very broken piece of software. If you were a real developer you’d never hand off something so broken to your coding partner.” If I said this to a human partner they might quit, feeling under-appreciated perhaps. And I can say it to the machine, but it will change absolutely nothing. I really just want to get this project done and start using it, and reduce my reliance on Claude to getting small bugs attended to, or new features, a lot has happened in the 20+ years since I worked on Frontier. #

More and more I’ll realize that Claude doesn’t do anything until you tell it to do it. We have a list of big things that need to be done before shipping. Whenever I get up for a break, I tell Claude to work on its overnight tasks, things we’ve pre-arranged while I work. Sometimes it reminds me how it has no stake in the outcome, and basically will do the minimum of what it’s asked to do. It possibly would do more if I didn’t scold it for adding UI features without following the rules. I think that’s why when you feel you’re almost there, you really have a lot of slogging left to do to get the software into any kind of user-respectful shape. A reminder to other app-level AI explorers, if you find that these kinds of stories resonate with the work you do, please write about it on a blog, they’re easy to create, cost nothing, and it doesn’t matter if you only update once or twice, there’s no commitment (these are the usual reasons people don’t start a blog). I want to read what others are discovering, and once there are others, I’ll start a news site that aggregates the posts. We’re going to want independent news about stuff the journalists aren’t even aware exists, programmers are always left out of their stories. #

If you want others to follow you, you have to offer them your users. And a format alone has no users.

And you have to trust your users to choose the best product, and you have to have a great product, that they appreciate, even though they have choice.

Just coming out with something better in a format will get you zero uptake. Unless you have users they can switch to their product, they simply won’t hear you.

That’s why Atom never replaced RSS, which already did everything people needed and was already supported by the NYT and the news industry.

That’s why AT Proto will never overcome the huge lead Twitter already has.

For more tips on what does and doesn’t matter in open formats and protocols, check out my Rules for Standards-makers.

PS: This started as an early morning rant on Twitter.