Another response, this time to a request from
@dela3499 for examples of working with AI tools.
My domain is automotive. Broadly. Too lazy to go into detail.
My approach is inefficient, but it's what works for me, at least so far.
If I understand correctly, they only "learn" within a context window, so I typically use Claude as a thinking partner. This develops thorough context.
I generate a working document throughout which details constraints, goals, and decisions made, including why those decisions were made.
I rarely accomplish this in one session, so the working document serves to provide the summarized content built from the last session. Most projects require 3 or 4 sessions.
When it comes time to code, Claude then has a clear picture, but I still have to make it present the architecture and get approval from me before generating code. Even with the context built from prior sessions, the first shot is almost always overly complex.
So I guess to sum up, I provide thorough context within a session, and then micromanage. It's much better at writing code than I am, but much worse at defining the architecture.
If you like, I will share a current working document. It's not top secret, but I don't want to post it for the entirety of X to see, so you'll have to send me a message.
Oh...one more thing. I have a knowledgebase which is just a bunch of .md files detailing various subjects. I often refer Claude to it so I don't have to type the same damn thing over and over.