Embarking on the journey of utilizing language models for coding, it seems the focus is shifting from perfecting one approach to exploring diverse workflows that balance pros and cons.
Embarking on the journey of utilizing language models for coding, it seems the focus is shifting from perfecting one approach to exploring diverse workflows that balance pros and cons.
Primarily, around 75% of the AI assistance in coding comes from tab completion, because demonstrating tasks through code chunks directly is more efficient than text communication.
It's about delivering task specifications succinctly, though sometimes the tab complete feature can be annoying and requires toggling.
The next step involves highlighting code sections for modifications. Further, tools like Claude Code or Codex are used alongside Cursor for larger functionalities, easily specified via prompts.
While helpful, they can veer off course, requiring frequent corrections and cleanups for style and clarity.
These tools often lack finesse, over-using try/catch statements, complicating abstractions, or duplicating code where helper functions would suffice.
They're particularly handy for less familiar coding areas or quick one-off utilities and debugging tasks, where creating temporary code is feasible without treating it as a finite resource.
For complex challenges, GPT-5 Pro is the go-to solution. After getting stuck with a bug, this tool often uncovers subtle issues after a thorough examination.
It excels in analyzing obscure documentation and providing insights on abstract solutions and literature reviews, offering substantial resources and guidance.
Overall, coding now offers vast possibilities with multiple tools, each with unique strengths and weaknesses. There's a blend of capabilities available, sparking curiosity about how others are navigating this rapidly evolving landscape.