How to Learn New Technologies: Take Them for a Drive
CTO Field Notes
SubscribeSign in
How to Learn New Technologies: Take Them for a Drive<br>A field note on the Driver, Mechanic, and Assembler approach to learning new technologies.
Joseph<br>Jun 09, 2026
Share
Over the last few years, I’ve changed the way I learn new technologies. Instead of starting by understanding how they work, I start by understanding what they can do. I’ve come to think of this as moving through three stages:<br>Driver,
Mechanic, and
Assembler.
The more technologies I encounter, the more convinced I am that most people should start as a Driver.<br>That wasn’t always my approach.<br>Over the last 30 years, I’ve lived through several technology waves: client-server systems, GUI programming, the internet, mobile, cloud, machine learning, and now GenAI. My instinct as an engineer was always to open the hood first.<br>If I was learning a GUI technology, I didn’t just want to know that window.alert() displayed a message. I wanted to understand exactly what happened after I typed it.<br>The same was true when I learned about the internet. I wanted to understand HTTP requests, DNS resolution, servers, application routing, and everything in between.<br>Curiosity served me well, but it also had a downside. I could spend a lot of time understanding how something worked before I understood whether it was actually useful.<br>GenAI is where I consciously flipped that model.<br>Driver
Think about learning to drive a car.<br>Most of us don’t begin by studying how fuel is converted into motion. We don’t start with the braking system, the steering linkage, or the transmission. We get into the car and drive it. We drive it in the city, on the highway, and up steep hills. We learn what it can do, where it performs well, and where it struggles.<br>Only after spending time behind the wheel we become interested in what’s happening under the hood.<br>That’s the approach I took with GenAI.<br>As soon as ChatGPT became available, I started using it. I used it to create policy documents for work. I used it to generate images. I used it to learn new subjects, research stock investing, rewrite essays, prepare sermons, and explore ideas. Whenever a new model appeared—ChatGPT, Claude, Perplexity, Mistral, and others—I took it for a drive.<br>At that stage, I wasn’t interested in model architectures, vector databases, or training techniques. I simply wanted to understand what these tools were good at and where they failed. I wanted to know what they could do for me.<br>The goal wasn’t to understand how the technology worked. The goal was to become proficient at using it. I wanted to master what the tool could do for me before I spent time learning what was happening under the hood.<br>Mechanic
After spending enough time as a Driver, curiosity naturally returned.<br>Now I wanted to know why certain things worked and why certain things didn’t. I started exploring APIs. I experimented with Open WebUI. I learned about system prompts, prompt storage, memory, context windows, and output processing.<br>What changed, however, was my intent.<br>Earlier in my career, I often felt the need to understand everything. With GenAI, I found myself learning only what was useful . Instead of studying the entire engine, I opened the hood to understand the parts that helped me get better results.<br>That distinction turned out to be important. I was no longer learning for the sake of completeness. I was learning to improve capability.<br>Assembler
Eventually, I found myself doing something different again.<br>I wasn’t just using AI tools, and I wasn’t just customizing them. I was starting to combine them to develop personal applications.<br>Claude helped me create requirements. I generated code via Google’s Antigravity. Cursor reviewed the code. I used LLM Models via Open Router. I also experimented with AI Agents. Instead of looking for a single tool that did everything, I started assembling workflows from multiple components.<br>This felt very similar to building a custom car. The engine comes from one place. The dashboard comes from another. The wheels come from somewhere else. Individually, they’re useful components. Together, they become a system.<br>Today, much of my experimentation with AI happens in this stage. I’m constantly trying different combinations of tools, models, prompts, and workflows to see what new capabilities emerge.<br>What Surprised Me
One of the things that surprised me most is that many non-technical people naturally start as Drivers.<br>Most people don’t care how a mobile phone works. They don’t need to understand radio frequencies, processors, operating systems, or network protocols. They care about what the phone helps them accomplish.<br>Because of that, they often adopt new technologies faster than engineers.<br>As technologists, we sometimes assume that understanding must come before usage. In practice, I’ve found that usage often creates the motivation for understanding. Once you see value, you have a reason to learn more...