Hello, everyone!
As you're all aware, we're currently part of one of the biggest shifts in software development since switching from machine code to high level languages. Agentic coding is moving forward in huge strides, and whatever one thinks about it – it's here to stay.
Starling, of course, is part of this new world, too. Yesterday, for the first time in this repository's history, I committed code that was basically created by Claude (golden image tests and a fix of the MultitextureStyle).
Over on GitHub, this immediately started a discussion about agentic coding. So I figured it would make sense to talk about this here, too.
I know that for many of us, agentic coding took away a significant part of the fun in our craft. I can totally relate – for example, I always appreciated having the privilege of writing APIs that other human beings are using, and I loved the feedback and the discussions we had about so many parts of it here or over at GitHub. In the future, new APIs will probably be read and used mainly by agents. That's something completely different.
However: personally, I am also awed by this technology, straight out of a science-fiction movie. Agentic coding can feel like a superpower, making code materialize so much faster than you could write it manually. Sometimes, it reminds me of the time when I started programming – being extremely excited about what was suddenly possible.
As with any new tool, it has a learning curve – it can write working code quickly, but it takes the right approach to keep quality standards up, and to not end up spending your days doing nothing but reviewing pages and pages of source code. Plus, it's a moving target, with new models and harnesses and skills being released at a crazy speed. That's why it can be overwhelming – to me, at least, it often feels like that.
Still: I like the challenge! The new 'Golden Images' feature is a good example of how I'm using agentic coding. The first attempt at the golden image code worked just fine – but it was just not elegant. I threw it away and gave specific instructions what to do better in the second attempt. After a few refactorings, I was happy. I reviewed every line, modified it and tinkered with it until I liked it.
For me, that's a great way to code. In the company I'm currently working in, I'm spending a lot of time optimizing our workflows and skills to make the best use of it. A lot of the "code" I'm writing is now Markdown, and instead of Texture Memory I have to juggle time resources of our team. Who would have thought that that would be what software development looks like in 2026? 🙉
How have your jobs changed in the last year? How do you approach this paradigm shift?