Some people get it.

They understand that to build with agentic, you... build with agentic.

This morning I sat with a friend and co-worker. He already built his own virtual assistant. Skills. Scripts. A context builder that files notes by person. He's miles ahead of most people I talk to about this.

Then he said the sentence.

"I'm going to take the transcript of our conversation and have my assistant turn it into a task list."

That is the whole trick to building with agentic.

Claude Code, Cursor, whatever you use. That's the harness. The model reasons. You talk to it. And yes, it is a conversation. Ask your assistant how to get things done without burning tokens like kindling.

I gave him three conversations to start with his VA. Start with the same three.

  • Talk to your VA about moving the memory out of markdown and into SQLite. A pile of files will drown the context window.
  • Talk to your VA about implementing loops. One agent hands work to the next, with a cap, until they agree.
  • Feed your VA ideas from outside its world. I use posts on X. You can use an Anthropic blog, an OpenAI post, a LinkedIn post like this one, whatever you find interesting. You do not have to understand it. You have to pass it on. See if your VA can learn from it.

I did not invent that list. I explained the situation to my chief of staff, Mike, and asked for advice.

My friend understood what it means to be agentic-first. He gets it.

Learn from him, go build your VA. Go improve your VA. Ask it how.

Originally posted on LinkedIn: "Ask it How"