Agentic AI is what takes your chatbot LLM (i.e. you asking ChatGPT questions and it giving you answers) to the next level – where it doesn't just give you answers, but it does things for you.
"How do I build a website?" (and AI answers you with steps to do so) becomes "Build this website for me." (and it does it for you).
For the stuff I'm building, I've used Lovable as an agentic AI coding. But let's take it a step higher.
I tried implementing OpenClaw but failed, so I decided to use Hermes instead. (AI nerds know what this stuff is, everyone else, read along for now lol.)
Hermes uses a large language model (in my case I'm using Claude Sonnet 4.6) as its brain, but it has a body and skills that can do things you ask it to do.
Like typical AI tinkerers, the first thing I asked Hermes to do was build email alerts for: stock trading opportunities, business ideas, real estate industry intelligence.
Then I asked it to replicate real estate websites I have built in Lovable, but to push the project to Vercel instead. (I'll talk about platform dependency issues in a later post.)
The easiest way to understand agentic AI is Jarvis, from Iron Man. Agentic AI is your personal assistant who doesn't just chat with you. It does stuff for you.
It's hard to imagine this actually happening if you've never used it. And there is a lot of technical stuff that happens behind the scenes in order for AI to actually do things for you.
I'm running Hermes on my gaming computer (no Mac Mini for me). I give it instructions via Telegram (so I can work with it wherever I am). It constantly upgrades its skillsets and remembers how I do things. So now it's time to build even more useful stuff.
But boy, it really eats up tokens.