First Post (Again)

Here we go. Again.

I've tried this before. Posting consistently about what I'm building. It never lasted more than four days. But this time something's different — and it's not just willpower.

What changed

I've been spending way more time on personal projects lately. The reason? AI agents handle a ton of the boring stuff now. Boilerplate, repetitive configs, scaffolding — I offload that and focus on the parts that actually require thinking.

But I'm not interested in just throwing prompts at these tools and hoping for the best. I want to understand how they work. What makes a good agent prompt vs a bad one? Why do some setups work and others loop forever? What's actually happening when an agent "reasons"?

What this blog is about

Me figuring things out and writing it up:

• Agents vs skills vs prompts — what works, what doesn't, what's overkill
• MCP, ACP and whatever new acronym shows up next week
• Papers I read, things I break, things I accidentally fix
• My actual commits turned into lessons — mistakes included

Mix of theory and practice. Mostly practice.

What's coming next

I'll be writing about how to structure agent prompts that don't produce garbage, the difference between an agent and a skill (it's not what you think), and how many of each you actually need before it becomes chaos.

Let's see if this one sticks.