Episode 33: FileMaker 2026 Review: More Than Meets the Eye

I

can.

Mine's

pretty quick.

So by and by large, this release to me felt like it was pretty much just setting up the next release.

A lot of things gearing up for more AI integration or having the AI be able to edit more.

So there isn't really a specific feature that you can use live in FileMaker that I'm super excited about.

But one thing that I do like, and I just started experimenting with, is with the FileMaker server installer, with the newest version, 26, you can install the command line tools only instead of the entire server package.

So you can install it on just a regular Windows machine.

I mean, technically you could do that before anyway, but it wasn't supported.

But now you can actually install the command line tools from the FileMaker server package directly on a Windows machine and experiment away with the updated pipeline to being able to edit FileMaker files with AI in the next release, hopefully.

Hopefully, yeah.

Yeah, that ties pretty tightly into some experimenting I was doing with the build file command line function that they introduced.

This is all in a blog post I wrote, so if anybody's read that blog post, I'll be repeating myself.

But I was watching a... a video stream that had Vince Manano and Hanzo in it from Two For You.

And one of them made a passing comment about maybe the build file function being a way of taking a file that had been recovered And then where the customer had continued to use it, to take that file, save as XML, use the build file function on the XML to get yourself back to a file that didn't violate FileMakeClaris' standing best practices, which is never to use a recovered file.

The best practice is if there's a file corruption, you grab your latest backup, You grab a clone, a known good clone, you export whatever data you can get out of the recovered file, and you import all the data into a known good clone or import it into the backup so that you're never using the recovered file.

As we, I think, all know, that's not what our clients are doing.

I doubt that very many of the inherited solutions we work with have been recovered less than a couple of times at least.

And so there's no way to have a known good solution.

you know, application in those cases.

And so anyway, that was some intriguing possibility.

And so I did a bunch of experimenting.

I have to say, you know, my old reactions to FileMaker, you know, like the... uh the migration tool some of the other tools we use all the time my old reaction to some of those would have been oh it's command lines they have to do it in command line now my reaction because of the ai revolution we're in now my reaction is oh you know a command line tool let's let's me tell tell the ai tell the large language model what to do with it and i don't have to really do any of the heavy heavy lifting and it's actually better i think So, um, I did a bunch of experimenting with, um, and it's in the blog post, so I'm not going to repeat myself entirely, but I did a whole bunch of experimenting with, with a large language model that shall not be named, um, and told it, um, you know, I want to do some practice testing.

So I, you know, I, I created a bunch of save XMLs from different solutions and had the, had the model do a bunch of, um, had the model do a bunch of work on them and, you know, analyze the results.