Honestly, I'm not mad about it. Most of my working hours now go into software where an LLM decides what to do next, calls the right tools, and drives a process from start to finish.

What that actually looks like

The work splits roughly into two flavors.

Wiring AI into existing systems. Taking software that already does its job and giving it a brain. That means automating the tedious steps where a human used to sit in the middle, and letting a model triage, classify, summarize, or kick off actions that used to need someone clicking through screens. It sounds fancier than it is. Most of the effort goes into boring edges like guardrails, fallbacks, and making sure the model can't confidently do something stupid.

Greenfield prototypes. Starting from an empty repository and asking what the product would look like if AI actions were the core instead of a bolt-on. This is the fun part, and the pace is no joke: the goal is usually a working proof of concept on real client data in the first week. You throw something together, learn that half of your assumptions were wrong, and the surviving half becomes the actual design.

What I've learned so far

The models are rarely the hard part anymore. The hard part is everything around them. How you shape the context, how you constrain the actions, how you recover when things go sideways, and how you make the whole thing observable enough to debug. Classic integration work, really: the interesting problems are always at the seams.

Where I want to go next

So far I have been on the integration side, using models rather than building them. I want to shift that balance a bit and dig into how an LLM actually works under the hood. No promises on timelines (see: this entire website), but that's the plan.

If you're tinkering with anything similar, always happy to compare notes. Reach out.

The name 'George' comes from when my wife and I first met online. I was being cagey about my real name, so she just called me 'George' on a whim. Her guess was so close that I gave up and told her my real name, which in Dutch translates directly to George. It stuck.

The handle BraveOtter came later, half as a joke and half to support my wife. A fresh start with no history attached. As for OtterSoftware: a pun on "Other Software", which says enough about the kind of things I build.

OtterSoftware started with a bigger plan. Build things live on stream, hang out with friends, grow a community of developers. Life had other plans. I landed a job at an AI development agency, where I focus fully on AI integration and automation, and streaming moved to the back burner.

I'm not complaining though. By trade I'm a full-stack developer, and these days my work is all about AI: building agents and automations for clients, wiring models into existing systems, and prototyping new ideas around them. Outside of work I still build odd things. Sometimes practical tools, sometimes projects that exist just for fun. Definitely not production-ready, but that's part of the charm.

So instead of a stream schedule, this site is where I write about whatever I'm building or thinking about, whenever I get around to it. If you want to see what I'm up to, my is the best place to look, and you can always reach out.

Thanks for stopping by!