I recently tried Vellum. Its is a Mac OS product but works pretty well. Ive found it keeps things much better organized than most AI assistants. Im having issues with TTS however.

Im trying to learn how to build my own Agentic system. I want a very capable non cloud AI Assistant.

Claude has probably made the biggest difference for me, especially for software planning and code review.

I try not to start by asking it to write an entire feature. I first give it the task and relevant project context, then ask it to produce an implementation plan or ADR. After I review the assumptions and edge cases, I let it generate the first version of the code and tests.

I also use skills from marketplace, search github or marketplace for relevant tools with high number of starts - increases productivity by 30%

The biggest productivity improvement is not faster typing. It is spending less time switching between requirements, architecture, implementation and testing. Human review is still necessary, especially when the original task contains missing or contradictory information. It is called HITL (human in the loop)