I’m busy all the time with the Critter Stack tools, answering questions on Slack or Discord, and trying like hell to make JasperFx Software go. I’ve admittedly had my head in the sand a bit about the AI tools for coding, thinking that what I do being relatively novel for the most part and that I wasn’t missing out on anything yet because the AI stuff was probably mostly trained up and useful for repetitive feature work.
The unfortunate analogy I have to make for myself is harking back to my first job as a piping engineer helping design big petrochemical plants. I got to work straight out of college with a fantastic team of senior engineers who were happy to teach me and to bring me along instead of just being dead weight for them. This just happened to be right at the time the larger company was transitioning from old fashioned paper blueprint drafting to 3D CAD models for the piping systems. Our team got a single high powered computer with a then revolutionary Riva 128 (with a gigantic 8 whole megabytes of memory!) video card that was powerful enough to let you zoom around the 3D models of the piping systems we were designing. Within a couple weeks I was much faster doing some kinds of common work than my older peers just because I knew how to use the new workstation tools to zip around the model of our piping systems. It occurred to me a couple weeks ago that in regards to AI I was probably on the wrong side of that earlier experience with 3D CAD models and knew it was time to take the plunge and get up to speed.
Anyway, enough of that. I spent a week thinking about what I’d try to do first with AI coding agents and spent some time watching some YouTube videos on writing prompts. I signed up for a Claude Max subscription at the beginning of last week to just go jump into the deep end. My tally so far in two weeks for progress is:
- Added MySql and Oracle database engine support to Weasel and Wolverine up to and including the ability for the Critter Stack to manage database migrations on the fly like we already did for PostgreSQL and SQL Server. Granted it took a couple attempts at the Oracle support, but it just doesn’t hurt to throwaway code that didn’t cost you much to write. Babu added Sqlite support as well.
- Filled in a gap in our SQL Server support for a queue per tenant database that had been outstanding for quite awhile
- I had Claude fix some holes in our compliance test suite for our RavenDb support I’d been neglecting for awhile
- Still in progress, but I have the beginning of a “Marten style migrations for EF Core” subsystem going that’s going to make the Wolverine testing for our EF Core integration go a lot smoother when the kinks are worked out as well as potentially making EF Core less aggravating to use for just about anyone
- I’m almost done with a potentially big performance optimization for Marten projections that I’d wanted to do for 6 months, but never had anywhere near enough time to research fully enough to do. That took in the end 30 minutes of my time and a couple hours of chugging. Just to make this point really hard here, it helps tremendously to have a large base of tests
- I improved quite a few “blinking” tests in the Wolverine codebase. Not perfect, but way better than before
- I pushed quite a few improvements to the Wolverine CI infrastructure. That’s a work in progress, but hey, it is progress
- I got a previously problematic test suite in Marten running in CI for the first time
- Marten’s open issue count (bugs and enhancements) is at 16 as I write this, and that’s the least that number has been since I filled out the initial story list in GitHub in late 2015.
- Wolverine’s open issue count is coincidentally down to 16. That number has hovered between 50-70 for the past several years. I was able to address a handful of LINQ related bugs that have been hanging around for years because the effort to reward ratios seemed all wrong
- I filled in some significant gaps in documentation in Wolverine that I’d been putting off for ages. I certainly went in after the fact and made edits, but we’re in better shape now. But of course, I’ve already got a tiny bit of feedback about something in that being wrong that I should have caught.
- I had Claude look for savings in object allocations in both Marten and Wolverine, and got plenty of little micro-optimizations – mostly around convenient usages of LINQ instead of slightly uglier C# usage. I’m not the very best guy in the world around low level things, so that’s been nice.
- I converted a couple of our solutions to centralized package management. That’s something I’ve kind of wanted to do for awhile, but who has time to mess with something like that in a big solution?
And really to make this sound a bit more impressive, this was with me doing 8 hours of workshops for a client and probably about 10-12 other meetings with clients during these two weeks so it’s not as if I had unbroken blocks of time in which to crank away. I also don’t have a terribly good handle on “Vibe Programming” and I’m not sure at all what a “Ralph Loop” is, so all of that very real progress was without me being completely up to speed on how to best incorporate AI tools.
Moreover, it’s already changed my perspective on the Critter Stack roadmap for this year because some things I’ve long wanted to do that sounded like too much work and too much risk now seem actually quite feasible based on the past couple weeks.
With all of that said, here are my general takeaways:
- I think Steve Yegge’s AI Vampire post is worth some thought — and I also just thought it was cool that Steve Yegge is still around because he has to be older than me. I think the usage of AI is a little exhausting sometimes just because it encourages you to do a lot of context shifting as you get long running AI agent work going on different codebases and different features.
- I already resent the feeling that I’m wasting time if I don’t have an agent loaded and churning
- It’s been great when you have very detailed compliance test frameworks that the AI tools can use to verify the completion of the work
- It’s also been great for tasks that have relatively straightforward acceptance criteria, but will involve a great deal of repetitive keystrokes to complete
- I’ve been completely shocked at how well Claude Opus has been able to pick up on some of the internal patterns within Marten and Wolverine and utilize them correctly in new features
- The Critter Stack community by and large does a great job of writing up reproduction steps and even reproduction GitHub repositories in bug reports. In many cases I’ve been able to say “suggest an approach to fix [link to github issue]” and been able to approve Claude’s suggestion.
- I’m still behind the learning curve, but a few times now I’ve gotten Claude to work interactively to explore approaches to new features and get to a point where I could just turn it loose
- Yeah, there’s no turning back unless the economic model falls apart
- I’m absolutely conflicted about tools like Claude clearly using *my* work and *my* writings in the public to build solutions that rival Marten and Wolverine and there’s already some cases of that happening
- The Tailwind thing upset me pretty badly truth be told
Anyway, I’m both horrified, elated, excited, and worried about the AI coding agents after just two weeks and I’m absolutely concerned about how that plays out in our industry, my own career, and our society.
The Shade Tree Developer 