I run my AI agents from an always-on mini PC. Ollama Cloud & Claude handle inference, mem0 keeps memory, & a simple workflow ties it together. Here is a post about my setup: https://giannis.32apps.com/blog/my-ai-setup/

A weekend off the keyboard and the AI feed looks like a decade passed by. I stopped tracking every announcement a while ago. Most of it is noise that disappears in a while. What actually holds up tends to find you sooner rather than later.

I tried paperclip.ai recently. It routes requests across multiple AI providers through one interface. The abstraction is clean & intuitive, you manage your "company" by creating agents. Very useful if you switch models often but not really worth it if just one provider covers your needs.

I was looking forward to using Kimi K3 model with ollama, but apparently you have to pay extra on top of your subscription. I understand it's a massive model that needs a lot of resources to run, but as a customer I see missed opportunity to get more subscribers on board.

I want to route different providers through Claude Code without juggling configs. Looking for a router tool that stays out of the way and just works. Something simple I can just plug & play, without having to do too much debugging.

I am not a designer. I let two AI tools handle the design of this blog instead of fumbling through it myself. Open Design & Claude Design both had their moments. I wrote up what worked and what did not. https://giannis.32apps.com/blog/how-i-designed-this-blog/

I get asked about persistent memory for AI agents often. I use mem0 and it has been solid. Claude-mem is a decent alternative if you work inside the Claude ecosystem as well.

I put up an intro post on my blog. It covers who I am, why the blog exists, & why it looks like a terminal from early 2000s. Check it out: https://giannis.32apps.com/blog/whoami/

I don't update dependencies by hand anymore. Bumping package versions & language runtimes are trivial tasks. An AI agent can handle them fine with the right instructions. I prefer spending that time on parts of the business that require actual thinking.

I kept everything in a single AWS account for years because it felt simpler. A contractor needed scoped access and a staging config bled into production. Separate accounts take longer to set up but they also keep a bad deploy from taking down everything else.

From to-do lists to DO lists. AI shifted the work from writing tasks to executing them. I stopped tracking things a model finishes before I finish typing. The interesting part is not the speed. It is deciding what still requires my judgment.

What is actually surprising is that the model announcements were a surprise at all. For years, we have been warned that China was catching up in AI. Yet the world is shocked when it starts to look like the moment may have arrived.

America needs to stop getting shocked by Chinese AI

Kimi K3 and Qwen3.8 should come as no surprise.

theverge.com

I’ve spent a lot of time building on AWS and working on AI systems, so I focus on what actually holds up in practice. Most of what I share on AWS, AI, startups, and tools comes from hands-on experience.