\nView Original Handwritten Notes\n
Original handwritten notes available
\n\n\n\n\n
\n\n\n\n
Working with a partner all day—whether human or not—is tiring. You really keep on going, and the pace is fast. Practically coding at the speed of thought, or somewhat slower. An aching back and body are the least of my concerns.
\n\n\n\n\n\n\n\n
It is truly hard to balance the day’s work, which has been much more productive in some senses, but also a bit jarring in terms of my overall flow and day. It really feels like everything has changed. And you can get sucked into code projects for days.
\n\n\n\n
So while you are building a new garden, you still have to tend the weeds, take out the trash, and do the routine.
\n\n\n\n
What it’s like—working with the machine. The back and forth, the commits, the branches. The waiting and watching and learning. The AI is incredibly creative as a software engineer. Much in the same way we would put together tools, reduce problems, try different angles and approaches. And the surprising additional thing it picks up or decides to check.
\n\n\n\n
And then there is—unknown line creativity? Maybe it is mastery. To fully know all languages of all utilities. And how might this design differ if it were ported and reimplemented from the bottom up in Python? And what then would be the best way to know which libraries to use, if any—and are there tools for that abstraction as well?
\n\n\n\n
Right now is the time to rate-limit out in the places that are exciting—both in terms of free services and paid, which tie directly into the foundational models. And if done well, like we are seeing with agents, editors—the tying together of a tool and giving it “agency” and the responsibility to execute on your behalf. And you are likewise responsible for what it creates and ensuring the changes contain nothing unsavory. There is a lot of trust put into something that could easily fashion a malicious script of ill repute, or something to simply wreak havoc, like random deletion of system files.
\n\n\n\n
There is much to think about, but I have found it most joyous to play and be creative—and leave the teaching to others for a bit, while I can show right now learning, use, and production are key.
\n\n\n\n
There is—is there?—a perfect tool for a more frequently interrupted, sometimes distracted lifestyle. The Sonnet 4 model has been incredible to work with. Using OpenAI 4.1 to create mock-up images toward product features to be—and like magic, it is off writing code, watching my action in the terminal, when every now and again I control+c the program and it then begins crunching away on the problem anew, either a code change or many, or maybe a test to better determine either a solution or how to reproduce a problem in a simpler setting.
\n\n\n\n
And things of such complexity as a Flutter app are able to be worked on so beautifully—and also conveniently fast to develop with hot reload. And I haven’t even gotten into debugging on that platform, which will become important.
\n\n\n\n
While not laws, there are a lot of things to learn while vibe coding—but fortunately, you can start and learn as you go.
\n