The Week I Couldn't Open a New Chat
The skill was never technical. It was there before the tools were, waiting for them to catch up.
Title: The Week I Could Not Open a New Chat
Subtitle: The skill was never technical. It was there before the tools were, waiting for them to catch up.
I had been using AI every day for weeks when I picked up a new tool and could not find the button to start a new conversation.
“How can I open a new chat? I do not see anything in my left hand menu?”
That same week, I ran an acquisition target through a ten-point evaluation framework and told the model to always drive me toward the next logical step.
Both of those are true. They happened days apart. The gap between them is the whole story.
The honest start
I am not a developer. I have never been the one writing the code.
For years I was the translator. The bridge between what customers actually wanted and what engineers had to build. I was fluent in the concepts. Systems, databases, cloud, how the pieces fit together. What I was not yet fluent in was these particular tools.
The early weeks were clumsy. I fought to get documents the model generated into Google Docs without the formatting falling apart. I worked out an interest-only payment on a small line of credit by hand because I did not trust the tool to do it right. I could not find the button to open a new chat.
And on the same screen, in the same week, I was setting up a holding company, evaluating businesses to buy, and briefing the model on my goals the way you would brief a new hire on their first morning. Here is who I am. Here is what I am building. Ask me ten questions so you can actually help.
The ambition was never the bottleneck. The tooling was.
The instinct
Here is the part I did not notice until I went back and read my own history.
From that first week, I opened every session the same way. Not “write me a thing.” Instead: recall everything we have discussed about this, confirm you understand it, and then I will give you instructions.
I was loading the context before I spent it. Treating memory as a resource to prime, not a slot machine to pull. I did that on day four, before I could reliably format a document.
Nobody taught me to do this. It just felt obvious.
That instinct, set the context then command, was the entire skill. Everything I built after that was the same move wired into something bigger.
What it became
Free no-code tools became a content engine. Prompts started carrying version numbers and changelogs like small products. One reply script grew into a scoring system, then into an automated platform running separate brand voices on its own. Then a real codebase. Then models orchestrated across multiple providers with a fallback if one went down.
Every step was the same instinct reaching for better infrastructure.
It helps to know where the instinct came from. Before any of this, I had run a regulated fund. I had sat on an institutional trading desk. I had calculated a fund’s high-water marks by hand and computed the fees off them. I already knew what operational plumbing costs when it breaks, because I had paid for it.
I was not learning to think in systems. I was finally holding tools that could keep up with how I already thought.
There is a moment in the logs where I tell the model to lock in the free tier as long as it has an API, because I cannot be doing any of this manually. A few weeks later I am asking how to make it execute a full plan without stopping to ask my permission at every step. That request, it turned out, already had a name. Someone had built claude --dangerously-skip-permissions for exactly the person I was becoming.
I was reaching for the thing before I knew what it was called.
The principle
So when people ask how to get good at this, I do not send them a list of prompts.
The skill was never technical. It is architectural. And it is older than any of the tools.
Point at the goal, not the task. Build the system, not the output. Load the context before you spend it. And verify the work against something real, because a confident answer and a correct answer are not the same thing.
The market taught me that last one years before AI existed. In markets, a wrong answer shows up in minutes. The screen goes red and you know. In slower systems the feedback is crueler. A model quietly mis-scoring who gets flagged, a denial that looks clean on a dashboard while the cost compounds for months before anyone notices.
Same lesson. Slower siren.
None of that requires knowing where the new-chat button is.
The tell
I keep that first exchange in mind. A person who understood systems for a living, stuck on where a menu item was. Not because it is embarrassing, but because it is the tell.
The button was never the problem. The person typing the question already knew exactly where he was going. He was just waiting for the tools to catch up.
So if you are somewhere near where I was, using this every day, sensing the leverage in it, not sure where to start building, the move is not to go learn the tools first. Open a conversation and tell it what you are actually trying to do. Treat it as a partner to think with, not a machine to operate. The tooling is the easy part. It will catch up to you faster than you expect.
I wrote a while back that leverage does not fix you. It reveals you. I meant it about capital and influence. It turns out it is just as true about a chat window.
Leverage was never going to make me into someone I was not. It only magnifies what is already there. The operator in that story was there the entire time. He was just waiting for something that could keep up.
So is yours.
-Jeremiah